Human Centric AI for Digital Accessibility I Axe-con 2026 — Transcript

Full transcript

1. 0:00Thank you, Glenda. Thank you for this wonderful  introduction and hello everybody from wherever
2. 0:05you are tuning in from. Um I'm very excited  and honored to be giving this keynote today.
3. 0:12So I've been on a mission to humanize technology  before it dehumanizes us. And um I really believe
4. 0:19human- centric AI is absolutely key for digital  accessibility. And um the good news is I also
5. 0:25believe we have the agency to make this  happen. Um so you know 45 years ago or so
6. 0:31um both Bill Gates and Steve Jobs posed this idea  of a personal computer for everyone and today over
7. 0:3980% of the world's population um actually have  access to a smart device which is amazing. Well I
8. 0:46believe we are on a um similar kind of trajectory  of giving a personal AI assistant for everyone. Um
9. 0:5357% of Americans today use AI for personal tasks  and over 40% of employees in the workforce um um
10. 1:03use AI for work. And um you know it's been now you  know 3 plus years since chat GPT was announced in
11. 1:13November 2022. Um and there are over 300 million  daily active users using chat GPT. And of course
12. 1:21there are other options for um AI tools uh that  function similarly as well. Um so what does
13. 1:28this world look like? I I believe this is very  exciting. AI is already ingrained in the fabric of
14. 1:33our everyday lives acting on our behalf. Perhaps  um you know you're asking it for health advice
15. 1:40or it's your health companion. Um perhaps it's  driving your car in a self-driving vehicle. Um
16. 1:46this is a picture of my son um many years ago um  interacting with a robot called Jibo which was an
17. 1:53MIT startup. Um there are many ways AI is deeply  ingrained in our in our lives and it is becoming
18. 1:59more and more mainstream every day. But here's  my concern. A lot of the AI we see today has high
19. 2:08cognitive intelligence, high IQ, but no empathy,  no emotional intelligence. Uh and if you think
20. 2:15about human intelligence, yes, your IQ matters,  but actually EQ, your emotional intelligence,
21. 2:22your emotional quotient, um your ability to  tap into how you're feeling and how others are
22. 2:28feeling and and act accordingly, that is actually  a better predictor for how um good a person is in
23. 2:35terms of leadership, in terms of being a friend  and a partner and a parent and um you know,
24. 2:41a a good community member. And so I am worried  that we are obsessed about the IQ of AI around us,
25. 2:48but we're not paying enough attention to the  social and emotional intelligence capabilities
26. 2:53of these technologies, especially that they're  so deeply ingrained in our everyday lives. And so
27. 3:00um I have been on this mission to bring emotional  intelligence to our machines. Um now what is very
28. 3:05interesting in AI is I do feel like we are um in  a somewhat dangerous territory. Um, yes, AI is
29. 3:12passing all these incredible again IQ benchmarks  that are mostly about IQ and productivity gains,
30. 3:19but who is what what's the equivalent benchmark  for um how AI is impacting um the social and moral
31. 3:27fabric of our society and very few people are  focused on that and I think that is a um that
32. 3:33is a concern. Um MIT's media lab just introduced  a program called AHA advancing humans with AI.
33. 3:41Um and they are partnered with the big AI labs to  try and build a benchmark that um that allows us
34. 3:47to quantify you know the safety of these models  and also um the impact of these models on human
35. 3:55relationships and perhaps even our mental health.  Right? We see a lot of um kind of really worrisome
36. 4:01headlines about how people are a using AI as as  companions but in a way that deters from their
37. 4:08um actual human relationships. Um we're also in  dangerous territory because um this this is this
38. 4:15is you know depending on the day and and the model  that's being released um you know you you we still
39. 4:22see examples of bias in these models. We still see  examples of disinformation and of course this has
40. 4:28been made worse with um um audio, video and image  generated AI and of course hallucinations which
41. 4:38which is kind of interesting because part of what  is interesting about these models is that they are
42. 4:44generative. So by definition they are supposed to  get creative. Um but but but often we see examples
43. 4:52where it's not at all re rooted in in in truth and  that is an issue. Um I as a parent I worry that
44. 5:02um this whole idea of artificial intimacy um is is  a concern especially for young people especially
45. 5:08for teenage boys. Um there are several platforms  that are geared towards building these companions
46. 5:16or therapist bots or AI friends towards young  people and um yeah you know Esther Pel who is
47. 5:28um kind of an expert on on relationships she calls  that a different type of AI artificial intimacy
48. 5:35and um Yuval Harrari says that the last decade was  about the race for human attention but the next
49. 5:42decade is a race for human intimacy and again  that's concern how are we building guard rails
50. 5:47into these systems so that I don't know if you're  a 14 14year-old and you are you've been chatting
51. 5:53with your AI friend for the last 3 hours maybe  it should actually say you know what like we're
52. 6:00calling it a day now go outside interact with your  family or your friends but of course some com you
53. 6:05know most companies wouldn't be incentivized  to do that because you want to maximize
54. 6:10time on the platform and engagement with these  bots because that's what is going to drive you
55. 6:16know revenue for the company but um I think the  right thing to do and that's what I look for when
56. 6:21I'm investing in startups the right thing to do  is to actually build guard rails around these
57. 6:25technologies to ensure safety um and of course  AI can exacerbate accessibility issues so um and
58. 6:34you know today 90 to 97% of websites still fail  to meet basic accessibility standards for users
59. 6:42with disabilities. Um 90 almost 97% of the top 1  million websites have uh web content accessibility
60. 6:51guideline failures. So they fail um these  basic again webs you know web contact content
61. 6:58accessibility guidelines um and and if you think  about it um a lot of um apps and websites today
62. 7:06are increasingly being um generated with AI so AI  coding um like VIP coding platforms like lovable
63. 7:14cursor um and others and if we are not which is  great because it accelerates the building of these
64. 7:22um you know digital assets. However, if we are not  generating this code with accessibility in mind,
65. 7:29it can very quickly make things worse.  So if accessibility is an afterthought,
66. 7:35it becomes really hard to um build build these  you know guidelines back in um after you've you
67. 7:43you know you've built your website and you  know in three hours kind of thing. So that
68. 7:48is a concern for sure and um I am personally very  passionate about um diversifying the face of AI
69. 7:56and who gets to build AI. I um I'm a true believer  that diversity of thought and perspective creates
70. 8:03better products and solutions for everyone. Um and  the more inclusive we are and the people who are
71. 8:09uh sitting around the table building and  ideulating and designing these technologies
72. 8:14that are going to influence every aspect of our  lives, the better we will be. And unfortunately
73. 8:19AI has a diversity problem. If you look at all the  lists of the top AI 100 companies or the top AI 50
74. 8:28companies or whatnot, you will quickly see that  there is a um uh es especially a gender diversity
75. 8:35uh issue. There's there's not enough women  building these um um yeah AI technologies that
76. 8:43we are using day in and day out now. uh and that  is a concern and and and again I try to um as as
77. 8:51an investor I try to seek uh incredible diverse  founders who are not just building differently
78. 8:57but they're solving different problems altogether  right because of their lived experiences and I
79. 9:02think that is so key um so I do think there's an  amazing opportunity it's still very early days in
80. 9:09AI and we again get like the seam this theme  of agency we get to shape how this all turns
81. 9:15out to be um and so I think the solution ought  to be humanentric AI. How do we center uh our
82. 9:22building of these AI technologies around what  you know what what humans uh need and I really
83. 9:29believe like many things in life we will not win  the digital accessibility challenge just with
84. 9:35um putting humans you know um at to work. We need  the scale of AI because this is how we're going
85. 9:42to accelerate bringing these solutions uh to the  world. Um and I also really believe that we need
86. 9:50a new social contract between humans and AI.  Um if you look at how um humans make decisions,
87. 9:57uh often it's rooted in empathy and it's rooted in  trust. And I and I do believe that we need to bake
88. 10:03that into the AI technologies that we use. We need  this new social contract between humans and AI.
89. 10:09Um, and I look for examples of that. Um, a little  bit about my journey. I started in academia. I
90. 10:14thought I'd become an academic. I did my PhD  in computer vision and machine learning uh at
91. 10:19Cambridge University. Um, my um, officemate Silus  Brown. Um, hi Silus if you're tuned into this. um
92. 10:28he uh is visually impaired and I I I learned a lot  of my um kind of experiences initial experiences
93. 10:38um kind of collaborating and co-working with  someone with a with with an impairment you
94. 10:45know the world is really not designed and  optimized for people with disabilities so
95. 10:51um and similarly um then I went to um MIT I was a  posttock at the MIT media lab And one of my really
96. 10:58kind of formative experiences was spending a  couple of days with a an individual on the autism
97. 11:04spectrum and realizing like, wow, the world  is so um overloaded with with with, you know,
98. 11:13stimulus. And it was just like the world was just  just existing in the world was overloading um this
99. 11:19person's uh sensory experiences. and um even just  navigating you know going from where this person
100. 11:27was staying to the MIT media lab it was just a lot  of um obstacles and again it made me realize that
101. 11:35um we are just we've designed the world in a way  that's not necessarily very inclusive. I then
102. 11:41started my company Affectiva. We were on a mission  to bring emotional intelligence to m to machines
103. 11:45and technology around us and then sold the company  to a Swedish publicly traded company called Smart
104. 11:52Eye. And I've now um shifted to um investing  investing in early stage AI startups that are
105. 11:59at the intersection of health and wellness, the  future of work and also sustainable living with
106. 12:04under this whole ethos of building humanentric  AI. Uh AI that unlocks human potential and a
107. 12:11amplifies and augments human ability, not replaces  it. Um when I first got to the uh um to MIT,
108. 12:18um we built these wearable glasses. This was,  you know, 2006, so wow, 20 years ago. Um,
109. 12:25we were building these glasses that were designed  to help individuals on the autism spectrum really
110. 12:31navigate the social and emotional um kind of  interactions um that they were part of. and we
111. 12:38collaborated very closely with a school in Rhode  Island uh called the Groden Center where we um
112. 12:44um kind of co-designed um in a very participatory  way these technologies with uh with the kids uh
113. 12:52at the school and I will circle back to this  use case at the end of my talk but it's it's
114. 12:57one of the most powerful use cases of emotion AI  in my opinion. Um we then um also um did a lot of
115. 13:06work in the automotive industry especially at my  company Affectiva and later at the company that
116. 13:11acquired a smart eye. And I just wanted to share  a few examples again of the power of how you can
117. 13:16use visual intelligence and computer vision and  AI to power um really amazing use cases in this
118. 13:23particular case. Um and I have permission to  share this video. Um this person is driving.
119. 13:30I'm going to play the video right now and you can  see there are four stages of drowsiness. He is
120. 13:37um quite asleep. He actually starts driving the  car at some point with um he's extremely tired and
121. 13:44sleepy. Um computer vision can fix this because  we can detect that his head is bobbing. His eyes
122. 13:49are closed and the car you can imagine how a car  can intervene in that moment and say you know what
123. 13:55you're very tired this is dangerous can't drive.  I'll take, you know, the car can take over driving
124. 14:00while he takes a power nap. Uh here is another  example. Again, we have permission. We actually
125. 14:08we gave these individuals cameras to install  in their vehicles and so they they knew that
126. 14:14they were being you kind of um that this data was  being collected essentially. Um I I I find that
127. 14:21um so she is texting while driving but not one  phone but two. So her her hands are off the wheel,
128. 14:30her eyes are off the road. Very easy for computer  vision to detect. And again you can imagine how
129. 14:36then you know um technology can intervene and  the car can intervene and take control while
130. 14:43she refocuses her attention. So we did a lot of  work for the automotive industry and there are a
131. 14:49lot of applications of this idea of understanding  your facial expressions, your voice intonations,
132. 14:56um your hand gestures, your body posture. This can  apply to not just the automotive industry but the
133. 15:02health um there are many many health use cases. Um  and um in today's age of AI, I truly believe truly
134. 15:10believe that this technology ought to be ingrained  in just how we interact with AI on a day-to-day
135. 15:15basis. ChatGpt ought to really understand  if you're stressed, if you're in a hurry,
136. 15:20um if you have time to chat, you know, um if  you're sad, if you're upset, like it should really
137. 15:26be able to take these nonverbal communications,  which by the way account for over 90% of how we
138. 15:32communicate on a day-to-day basis. Um, so I  wanted to kind of rope in like some of the
139. 15:39key innovations that I believe can really unlock  digital accessibility. And of course, the first
140. 15:44one is agentic AI. And and I just want to make  the distinction between AI bots and AI agents. Um,
141. 15:50AI bots are I'm sure we've all at this point have  had experiences interacting with uh AI chatbots.
142. 15:57So these are um kind of AI interfaces where you  could ask questions whether in text or voice.
143. 16:03um and you you know it's you you you can be  in a conversation with this bot but it doesn't
144. 16:09necessarily go execute anything on your behalf. an  AI agent um can actually execute um on your behalf
145. 16:18and uh we're already seeing the rise of AI agents  definitely in enterprises and increasingly for
146. 16:24the consumer uh for consumers um and I think that  presents a huge opportunity for the accessibility
147. 16:32um world because can you imagine if um um all  these AI generated coding platforms like cursor
148. 16:43like lovable actually adhere to accessibility  guidelines. So every time you know someone in
149. 16:51Egypt where I'm originally from decides to build  a web app or a phone app um and um is generating
150. 17:00this code like often often people who are vibe  coding aren't even technical right so um it's kind
151. 17:06of hard to imagine the you know these individuals  building in accessibility but what if it's already
152. 17:13baked into these um code generator platforms terms  um and every time you build a website or you build
153. 17:21um you you generate an app you've you've already  built in these accessibility guidelines. I think
154. 17:26that's a huge opportunity. Um I think there's  an opportunity to also integrate validation
155. 17:32frameworks and benchmarks like DQ's benchmarks  directly into the AI development workflow. So
156. 17:38again it's not an afterthought. Um, I also believe  that, um, we're going to be increasingly seeing
157. 17:45personal AI agents that not just act on your  behalf, but can personalize um, the experience,
158. 17:52can personalize the interface, can personalize  its recommendations based on your needs because
159. 17:57it knows you really well and it knows the best way  to interface with you. I'm very visual. I don't
160. 18:04um I struggle with like like long texts and I  would love if my AI knew that and it would kind of
161. 18:12represent things and mind maps and flowcharts and  and and visual methods. It doesn't do that yet,
162. 18:18but I believe we will get to the point where  um yeah, where that is a possibility. And then
163. 18:24lastly, I believe there's an opportunity for more  of what we call guardian AIs. um AIs that actually
164. 18:31monitor what the AI agents do and can um again  can kind of enforce guard rails can flag when
165. 18:38something is perhaps inaccessible or non-inclusive  or biased etc. Um, and we're starting to see the
166. 18:45rise of these guardian AI, so AIs that monitor  other AIs as well. Um, you may have all um seen
167. 18:52the news around or maybe even used Open Claw. Um,  this is an open-source AI assistant that kind of
168. 19:00had a had its viral moment a couple of weeks  ago. Um what was cool about it, it was um the
169. 19:06first instance really of an autonomous proactive  a AI agent that we've seen consumers use. Again,
170. 19:12there's a lot of use cases of AI agents in the  enterprise world, but not necessarily in um in
171. 19:19the um for consumers. And this was an example  where you could download this agent and it could
172. 19:26triage your inbox, manage your calendar, file  your expenses, book your doctor's appointments,
173. 19:31um do work on your behalf. And um what was really  fascinating was uh somebody built Maltbook on top
174. 19:39of it, which is the it's like a social network  for AI agents. You literally have to disclose if
175. 19:44you're getting as you kind of click on the  website if you're a human or an AI agent.
176. 19:49um because only the AI agents are allowed to  interact with each other and they ask each other
177. 19:54questions, they discuss topics, they upvote each  other. Um and it's the first time we're seeing
178. 20:01multi- aent communication at scale. Um I will say  this is exciting. It's directionally correct, but
179. 20:08um there are major major concerns in how this  particular instance of AI agents for consumers
180. 20:14was implemented. There was almost little zero  to no human um oversight. Uh there um was no
181. 20:23security implemented in these uh agents. Um  so so they were very susceptible to prompt
182. 20:30injection to cross agent manipulation, data leaks,  malware and scams. Um and so I've personally not
183. 20:37um used open claw but we do use cloud code which  is anthropics um um anthropics coding kind of AI
184. 20:46agent and it's been really interesting. It's been  quite powerful. Um, so I, you know, it's still I
185. 20:53think there is still a barrier to usage if you're  not super technical, but we're uh with every with
186. 21:00every release of these models, I believe we are  getting closer and closer to making this type of
187. 21:04technology available to consumers, which will be  very exciting. All right, innovation number two,
188. 21:10AI native devices. So, um, you know, your  smartphone is not an AI native device. Um,
189. 21:16um, and I believe we are going to start seeing  AI native devices soon. So these are devices that
190. 21:23have you know implement seamless human machine  interaction um by being multimodal. So it's not
191. 21:29just textbased, it's not just voice based. Perhaps  it has perception, it has computer vision, it is
192. 21:35context aware, it has memory and it has empathy.  Um the form factor is still evolving. There are
193. 21:41uh companies experimenting with uh wearable  glasses. Um smart glasses obviously meta is
194. 21:49um kind of I I think perhaps the leader there.  um they've invested um three and a half billion
195. 21:56dollars in Rayban and um and and they're creating  these smart AI glasses. Uh I think what is very
196. 22:03interesting, we were my son and I my son is 17 and  I we were both at a tech conference and we were
197. 22:09given these meta glasses as gifts and I tried them  on. I was like ah not for me. My son though loves
198. 22:14them and he uses them all the time when we're  traveling especially he'll put the glasses on.
199. 22:19He'll ask for questions on what these landmarks  are and um he'll play music, he'll record,
200. 22:26take pictures. It's quite fascinating to see how  comfortable he is with uh with these glasses.
201. 22:32And then our um the school he's at, um there's  basically a debate because kids are increasingly
202. 22:40coming into class with these smart glasses. And  so in the same way that the school had to come up
203. 22:46with um um a point of view on cell phone usage,  um they're now having to come up with a point of
204. 22:54view on is it acceptable to be on campus with your  smart glasses? Can you show up to class with your
205. 23:01smart glasses that can take notes on your behalf?  Right? Are you allowed to have an AI notetaker
206. 23:06in class? So um I I think as this kind of next AI  native interface um starts to take shape, we will
207. 23:14we we will have to grapple with what is acceptable  and what's not. Especially as these devices can be
208. 23:20always on. They can always be recording. They  could always be taking pictures and videos and
209. 23:24notes, which on the one hand can be exciting. Um,  and I'll talk about some of the kind of what this
210. 23:31means for the accessibility world, but I I think  there's a lot of question marks around privacy and
211. 23:37whatnot. Open AAI is of course partnered with  Johnny IV, who was one of the designers of the
212. 23:43iPhone to also create an AI device. There's a lot  of chatter about what this is going to look like.
213. 23:48Is it going to be a pin? Is it an earbud? Is it  something ambient that you put in your, you know,
214. 23:53your space? TBD on that. Um the most important  thing though about these AI native devices is that
215. 24:00they are going to have to remember your context  and act proactively on your behalf. Um so memory
216. 24:06plays a very important um part in that too. Um  I really believe that to unlock to truly unlock
217. 24:15um the power of these kind of AI native devices  you will I believe we have to kind of marry human
218. 24:21intelligence with artificial intelligence. And  two particular examples um that I love and Glenda,
219. 24:27thank you for pointing me to those is the first  is Be My Eyes. And that is an basically a phone
220. 24:34app where you can bring in live interpreters  and live um assistants who can be your eyes.
221. 24:41And they just literally just um a few days ago  announced a partnership with Amtrak to expand
222. 24:48this live visual support to uh to over 50 stations  nationwide. as you're navigating um the station
223. 24:55um you can you know you can have uh people help  you navigate that remotely right virtually and
224. 25:02remotely which is very powerful and then there are  a whole slew of smart glasses an example is IRA
225. 25:08u very similar to Meta's glasses where you've  got these cameras that are looking outward
226. 25:15um that can again be your eyes um sometimes you  can bring in a human to help you navigate but uh
227. 25:22increasingly you can uh power that with artificial  intelligence with visual intelligence. A lot of
228. 25:28these models include visual understanding and  we're starting to see more and more of that. So
229. 25:33very exciting. The third innovation is physical  AI. AI is leaving the screen and going into the
230. 25:40physical world. So what does accessibility  look like in that world? Um there are a few
231. 25:45examples here that I've highlighted. Not quite  the Jetsons yet, but obviously um Elon Musk has
232. 25:52his Optimus robot figure AI um continue to  to to raise uh lots and lots of uh funding
233. 26:00to build their humanoid robot. and physical  intelligence um is also um kind of building these
234. 26:08um these these are all robots that are by the way  designed to to be in um in industrial situations
235. 26:16but also the home and I'm sitting here thinking I  don't want this guy like in my house that is and
236. 26:22again this is where I think building with empathy  and emotional intelligence in mind is so key if
237. 26:28we're expecting these devices to work along work  and live alongside with us then um uh then then
238. 26:36then we need to design these things differently in  my opinion. Um yeah and as you can see this robot
239. 26:44is really really really struggling with uh folding  laundry. Um it will eventually get there but we
240. 26:50don't have time for that. Um of course it is so  important to build with ethics and safety in mind.
241. 26:58Um, I believe that it's so important to um kind  of prioritize the ethical development and the
242. 27:05ethical deployment of AI. The ethical development  of AI is all about thinking about and prioritizing
243. 27:12bias from the get-go. Again, if you do this as  an afterthought, so you build your AI solution,
244. 27:18whatever it is, your AI product, and then at the  very end, just before you ship it, you're like,
245. 27:22"Oh, is there bias in here? Is the data, you know,  is there data or algorithmic bias? is um um is
246. 27:30this accessible? Is this inclusive? It's too late.  It has to be baked in to the beginning like from
247. 27:36the get-go from the throughout the entire design  and development process. Um and at my company,
248. 27:42Affectiva, we were strickers for that. in fact  a CEO of the company to ensure that this was a
249. 27:48priority because we were a small startup and  we were always kind of under the gun to ship
250. 27:53um products um right like as a small startup to  stay competitive but to ensure that we prioritize
251. 28:01the ethical development of AI um I tied everybody  like especially my executive team's um kind of
252. 28:10bonuses and and and compensation I tied that not  only to driving revenue um and profitability,
253. 28:18but also are we adhering to our standard, you  know, we we were we kind of were very we were
254. 28:25pioneers in in kind of articulating what an  ethical development framework looks like and
255. 28:31um yeah, our compensation was tied to to it and  and guess what that set a very strong tone that
256. 28:38this matters. We are going to evaluate these  models not just on how well they perform from
257. 28:43an accuracy standpoint but also is there any  bias that has creeped into these models and then
258. 28:49at the same time the the deployment of AI is very  important and and again at my company Affectiva we
259. 28:58knew that there were specific applications of this  technology that did not respect people's privacy
260. 29:04where you don't have an opportunity to opt in or  opt out of this. So, for example, we decided very
261. 29:10early on that we were not going to deploy this  technology in public spaces um because people
262. 29:16were often not given an opportunity to opt out of  this. Um and so um that that meant that there was
263. 29:23a whole slew of industries and applications and  use cases that we were not going to entertain as a
264. 29:29company which probably meant that overall we drove  less revenue but at the same time I feel like we
265. 29:36um demonstrated a a great we set the bar high for  what what where and what this technology should be
266. 29:45used for. And as an investor, I apply the same  lens when I'm looking at companies. So if I am
267. 29:52interviewing, you know, I'm in a conversation  with founders and trying to evaluate if I want
268. 29:56to invest and they have not thought at all about  ethics, they have not thought at all about safety.
269. 30:01If they, you know, if if we ask about like  whether they've taste whether they've tested
270. 30:08for the safety of of some of these models and they  haven't, then that's that's a red flag. Um, and I
271. 30:14do want to give a shout out to my dear friend Eric  Cohen, who is building an AI safety framework,
272. 30:20um, especially around mental health, um, and how  do we build guard rails into some of these models,
273. 30:27um, to ensure that it's not giving, you know,  misleading and really potentially very dangerous
274. 30:35advice to people um, around self harm, etc. And  so, um, I encourage you to check that framework
275. 30:41out. Um but yeah, it is so important to prioritize  kind of these elements of how are we building this
276. 30:51technology and where are we choosing to deploy  it? Where are we prioritizing our money, our
277. 30:57mind mind share um and and I think again we have  agency. We I I believe we have agency in terms of
278. 31:04how we build it, how we use it and and where we  deploy these technologies. Um, I always like to
279. 31:11end with this next video. Um, tying it back to  autism again. Um, we partnered with a company
280. 31:18called Brain Power. Um, and again it was kind  of interesting for our little startup Affectiva
281. 31:23when we decided to do that because, um, Brain  Power was very focused on an autism use case,
282. 31:30which at the time a lot of investors would tell  us, well, that's not that's a very niche market.
283. 31:35It's not, you know, it's not like this huge market  opportunity or TAM, right? The total addressable
284. 31:40market isn't that big, but we believe very  strongly that we um it's it's an incredible use
285. 31:46case. I just want to share um how they partnered  with us um to build assistive technologies for
286. 31:54um kids on the autism spectrum. I will say um the  video is a little quiet. I apologize for that. Um,
287. 32:02but there is there are captions at the bottom  as well. What do you see on screen? Mom.
288. 32:09Eight-year-old Matthew Kger has been diagnosed  with autism. A lot of the trouble he gets
289. 32:13into with other kids is he thinks he's funny and  doesn't read at all. That he's not or that they're
290. 32:18annoyed or angry. Matthew's mother, Laura, signed  him up for a clinical trial being conducted by
291. 32:24Ned Saheen. I want to know what's going on inside  the the brain of someone with autism. And it turns
292. 32:32out parents want to know that too. You get points  for looking for a while and then even for looking
293. 32:38away and then looking back. Saheen's company Brain  Power uses Affectiva software in programs Matthew
294. 32:44sees through Google Glass. These games are trying  to help him understand how facial expressions
295. 32:49correspond to emotions and learn social cues.  One of the key life skills is understanding
296. 32:55the emotions of others. And another is looking in  their direction when they're speaking. Looking at
297. 33:02at your mom and while it's green, you're getting  points. When it starts to get orange and red,
298. 33:07you're you'll slow down with the points. Am I  looking at you? You are looking at Just a few
299. 33:13minutes later, the difference in Matthew's gaze  overwhelmed his mother. I'm going to cry. Why?
300. 33:23Cuz when you look at me,
301. 33:29it makes me think we haven't really before  cuz you're looking at me differently.
302. 33:36So we know um we we we we we know from how um from  the results that Brain Power is getting um that
303. 33:47um the the the kids who are using this assisted  technology are actually getting better in terms of
304. 33:54how they are making eye contact, how often they're  making eye contact and um their understanding of
305. 34:00these facial expressions and facial cues. So  it's really powerful. The key question that the
306. 34:04company's trying to answer is if you take away  the glasses, does this learning generalize? U
307. 34:10which would be of course incredibly incredibly  impactful. Um so um I think just to wrap up um
308. 34:19one of the things that I'm really committed to is  amplifying the voices in AI that you often do not
309. 34:25hear from. Um and so about a year ago I started a  podcast called Pioneers of AI. It's a weekly show.
310. 34:32It's both on audio and video and we try to  of course host some of the big AI thinkers
311. 34:39that you often hear from in the AI headlines, but  I'm also committed to finding and amplifying and
312. 34:47platforming voices in AI that you don't hear  from. So, we've had incredible founders and
313. 34:52incredible builders in this space um that don't  get a lot of attention, but I believe are building
314. 34:58um some incredible applications of AI in  a very thoughtful and intentional way. So,
315. 35:03um so please please please check it out and let  me know what you think. And also, if you have
316. 35:08suggestions of um amazing people we should have on  the podcast, uh please let me know as well. Um I'm
317. 35:14very easy to find online and um yeah would love  to stay in touch with you all. Thank you so much.
318. 35:25Absolutely wonderful Ra. Thanks so much for  kicking us off and the chat and Q&A over in
319. 35:34Pubble has been on fire. Okay, that's great.  Wonderful. Let's bring it on. Exactly. Um,
320. 35:43the most upvoted question. Are you ready? Um, Ros  asks, "There's still huge amount of concern about
321. 35:52environmental sustainability of all this news  reporting that Texans have areas where there is
322. 35:58not enough drinking water because it is all being  used for cooling. How can we justify continuing
323. 36:05with this on a widespread basis when there is  such a threat to the world we live in?" Yes. Um
324. 36:12I um we've actually had I'm just going to point  to two incredible um um guests that we've had on
325. 36:19the show. Uh one is Sasha do Dr. Sasha Luchion.  She's on the Time AI 100 list and she is the
326. 36:27climate lead at this AI company called Hugging  Face. um some of you might be familiar with it
327. 36:33and her whole thing is we need more benchmarks um  around sustainability and about the environmental
328. 36:39impact of these AI models. So she's on this  mission again back to like how are we testing
329. 36:46these models right now we're just obsessed  about the IQ of these models but there are
330. 36:52other factors we need to be considering and what  environment is one of them. So she's awesome and
331. 36:56then Dr. Professor Priya Danti at MIT. Um she's um  spearheading using AI to be part of the solution,
332. 37:03not just the problem. And so I would encourage you  and she's also about inclusion and advocacy and um
333. 37:11how do we democratize Yeah. Dem democratize access  to these models in a way that is sustainable.
334. 37:19So I would encourage you to check her work out.  Definitely an issue. Absolutely. Absolutely. Um,
335. 37:26Tony asks, "Do we have any answer to the legal  questions that have an AI agent acting on your
336. 37:34behalf? Who's responsible for the action it  takes? Is it responsible when it makes the
337. 37:41mistakes?" That is a great question, Tony. Um,  accountability is so unclear in this world. Is
338. 37:48it? Um so for example say you are a lawyer at  a big law firm and you're using one of these
339. 37:54um AI tools um that the company you know the  company basically has a license to one of the
340. 38:01AI legal firms and something goes wrong. Is that  on you because you use it incorrectly? Is that on
341. 38:09the law firm because the you know it's the law  firm that's using these or is it the builder of
342. 38:15the technology? Right? Who's accountable? And  we right now we're in a very gray zone. It's
343. 38:20very unclear who's held accountable and I we need  better accountability frameworks. I also believe
344. 38:27regulation can be part of this um thoughtful  regulation. Um but regulation is always a few
345. 38:34steps behind. So ideally in an ideal world we hold  the bar high like we as in the professionals who
346. 38:40are building and deploying these technologies and  using them. Um, I'm a big proponent of like, yeah,
347. 38:48we need to set the bar high and decide what that  what this accountability framework looks like.
348. 38:54Absolutely. Here's another one. Jill asks, "How  can we motivate Vibe computing coding companies
349. 39:04to create WCAG compliant solutions?" Yeah.  Um I I think it's one of these examples where
350. 39:13um doing good is good for business, right? Like  um I I I I think it's a matter of I don't think
351. 39:24it's easy, by the way. Um I' I've been trying  to get the attention of one of these vibe coding
352. 39:29platforms to to to broke broker this connection  and it's it's you know, I haven't heard back yet,
353. 39:35but I'm not going to give up. Um but I but I  think it's worth exploring and at the end of
354. 39:41the day all you need I always used to tell  my my my team that all you need is is one
355. 39:46champion within these companies who are going  to take this on and be passionate about it. So
356. 39:51I I think it's a fight worth fighting and again I  think it's going to be good for these VIP coding
357. 39:56platforms if they build this from the get-go  if they build accessibility guidelines from
358. 40:00the get-go. Um so so yeah. Awesome. Awesome.  All right, let's see what's next. Alex asks,
359. 40:12"How might we manage our cognitive loads together  as humans in diverse spheres of agency?" Oo,
360. 40:21that's a hard question. You know, one one of  my predictions Yeah. Yeah. Alex, one of my
361. 40:29predictions for this year was that AI is going  to become more collaborative. So if you think
362. 40:36about most of the AI we see today, it's basically  one-on-one. I have my chat GPT, you have your chat
363. 40:42GPT instance, right? or even at at my fund Blue  Tulip Adventures, we're experimenting with um clot
364. 40:49code and we have our own like chief of staff AI  agent which has been really fascinating to stand
365. 40:56it up and give it access to all of my email and  my calendar and my LinkedIn and but it but still
366. 41:04like it's very hard to collaborate with AI at the  moment and I believe that's going to be the next
367. 41:11frontier of innovation like how do we build AI  as a co-orker as um yeah as a collaborator and
368. 41:22what does that what does that framework look like  right and and h how do we delegate and actually
369. 41:27um this is this is a great show it's it's  called Shellgame season 2 um Evan Ratliff
370. 41:36decided to start a company with two co-founders  that are AI agents literally and He made one
371. 41:43of the AI agents the CEO. So Kyle Law is an is  the CEO of this company, Humuro Hurimo, sorry,
372. 41:50Hur Humo AI. And but Kyle is an AI agent and I  actually got to speak to Kyle on Zoom. Like I got
373. 41:58to interview Kyle and he's like your typical tech  bro. And the COO Megan is also an AI agent. So he
374. 42:07was exploring this version of the world where we  live and work in hybrid AI human environments. And
375. 42:13we're not there yet, but he was really pushing  the boundaries of delegation and what happens
376. 42:19when an AI messes up, right? How do you bring in  other humans in the loop? Like there's a whole
377. 42:24episode on how they were trying to hire a human  product marketing manager and she realized that
378. 42:30her boss is an AI agent. That was interesting,  right? So, um, we that I I think what is kind
379. 42:38of anxietyinducing a little bit is like we're  in this world where there are no answers. It's
380. 42:44h it's unfolding. We're in it and it's unfolding  in real time. But it also makes it really exciting
381. 42:50and gives us agency. It does. It does. There are  so many good questions and Alex asked another one,
382. 42:57but I just asked one from Alex. Okay, I'm gonna go  with Alex's next question. Y'all forgive me. Um,
383. 43:06it's an important question. How might we avoid  the notion that someone that has autism needs to
384. 43:13be fixed? Yeah, I again I I spent a lot of time  in interacting with individuals on the autism
385. 43:23spectrum and there's no fixing. It's just it's  it's a diff it's a different way of perceiving
386. 43:30and being in the world. And sometimes actually can  be really powerful like some of some of yeah some
387. 43:37of the individuals on the spectrum like wow  like their cognitive abilities their pattern
388. 43:42recognition how they can stay focused like there  are so many things that are superpowers but it's
389. 43:48just unfortunately how we view the world how we've  designed the de facto like the default um default
390. 43:55way of being in the world that is that makes it  so hard. So my hope is um and that that that's
391. 44:02kind of the spirit of the research I did back  at MIT is that we yeah we find a way to to bring
392. 44:09empathy into this right to have empathy and and be  inclusive and yeah I I think empathy is a key part
393. 44:16of the solution honestly. Yeah I I couldn't agree  with you more. Um, I think embracing diversity
394. 44:25and recognizing that different can be absolutely  wonderful. Well, Ra, I think we're um about one
395. 44:37minute out from our end time and so many questions  that we didn't get to, but I want to thank you
396. 44:46so much. She reminded us that the future of  AI is not something that's simply happening
397. 44:51to us. It's something we can shape and that  accessibility is not a constraint on innovation.
398. 45:00It really is how innovation can serve humanity  and that human centric AI ultimately comes down
399. 45:07to choice about agency, dignity and whose guide  our designs. So, as we continue through Axe-con,
400. 45:18I invite all of you to carry these thoughts  forward. How can what we build next expand
401. 45:26possibility for more people? Thank you again,  RA, for opening Axe-con 2026 with such clarity,
402. 45:34care, and vision. And to the rest of y'all, go  enjoy your Axecon. We are just getting started.