Why Are People Starting to Sound Like ChatGPT? | Adam Aleksic | TED — Transcript

Full transcript

1. 0:04How sure are you that you can tell what's real online?
2. 0:08(Laughter)
3. 0:09You might think it's easy to spot an obviously AI-generated image,
4. 0:12and you're probably aware that algorithms are biased in some way.
5. 0:16But all the evidence
6. 0:17is suggesting that we're pretty bad at understanding that
7. 0:20on a subconscious level.
8. 0:21Take, for example, the growing perception gap in America.
9. 0:23We keep over- and overestimating
10. 0:25how extreme other people's political beliefs are,
11. 0:28and this is only getting worse with social media,
12. 0:30because algorithms show us the most extreme picture of reality.
13. 0:33As an etymologist and content creator,
14. 0:35I always see controversial messages go more viral
15. 0:38because they generate more engagement than a neutral perspective.
16. 0:41But that means we all end up seeing this more extreme version of reality,
17. 0:45and we're clearly starting to confuse that with actual reality.
18. 0:49The same thing is currently happening with AI chatbots,
19. 0:52because you probably assume that ChatGPT is speaking English to you,
20. 0:55except it's not speaking English,
21. 0:57in the same way that the algorithm's not showing you reality.
22. 1:00There are always distortions,
23. 1:01depending on what goes into the model and how it's trained.
24. 1:04Like we know that ChatGPT says “delve” at way higher rates than usual,
25. 1:08possibly because OpenAI outsourced its training process to workers in Nigeria
26. 1:11who do, actually, say, "delve" more frequently.
27. 1:14Over time, though, that little linguistic overrepresentation
28. 1:17got reinforced into the model even more than in the workers' own dialects.
29. 1:20Now that's affecting everybody's language.
30. 1:22Multiple studies have found that, since ChatGPT came out,
31. 1:25people everywhere have been saying the word "delve" more
32. 1:28in spontaneous spoken conversation.
33. 1:30Essentially, we're subconsciously confusing the AI version of language
34. 1:33with actual language.
35. 1:35But that means that the real thing is, ironically, getting closer
36. 1:38to the machine version of the thing.
37. 1:40We're in a positive feedback loop with the AI representing reality,
38. 1:43us thinking that's the real reality,
39. 1:45and regurgitating it so the AI can be fed more of our data.
40. 1:48You can also see this with the algorithm through words like "hyperpop,"
41. 1:51[not a] part of our cultural lexicon until Spotify noticed an emerging cluster
42. 1:55of similar users in their algorithm.
43. 1:57[When] they identified it and introduced a hyperpop playlist, however,
44. 2:01the aesthetic was given a direction.
45. 2:03Now people began to debate what did and did not qualify as hyperpop.
46. 2:07The label and the playlist made the phenomenon more real
47. 2:09by giving them something to identify with or against.
48. 2:12And as more people identified with hyperpop,
49. 2:14more musicians also started making hyperpop music.
50. 2:18All the while, the cluster of similar listeners in the algorithm
51. 2:21grew larger, and Spotify kept pushing it more,
52. 2:24because these platforms want to amplify cultural trends to keep you on the app.
53. 2:29But that means we also lose the distinction between a real trend
54. 2:32and an artificially inflated trend.
55. 2:34And yet, this is how all fads now enter the mainstream.
56. 2:38We start with a latent cultural desire.
57. 2:39Maybe some people are interested in matcha, Labubu or Dubai chocolate.
58. 2:43The algorithm identifies this desire and pushes it to similar users,
59. 2:46making the phenomenon more of a thing.
60. 2:48But again, just like how ChatGPT misrepresented the word "delve,"
61. 2:52the algorithm is probably misrepresenting reality.
62. 2:54Now more businesses are making Labubu content
63. 2:56because they think that's the desire.
64. 2:59More influencers are also making Labubu trends
65. 3:01because we have to tap into trends to go viral.
66. 3:04And yet, the algorithm is only showing you
67. 3:06the visually provocative items that work in the video format.
68. 3:11TikTok has a limited idea of who you are as a user,
69. 3:14and there's no way that matches up
70. 3:16with your complex desires as a human being.
71. 3:18So we have a biased input.
72. 3:20And that's assuming that social media is trying
73. 3:22to faithfully represent reality, which it isn't.
74. 3:24It's only trying to do what's going to make money for them.
75. 3:27It's in Spotify's interest to have you listening to hyperpop,
76. 3:30and it’s in TikTok’s to have you looking at Labubus
77. 3:32because that's commodifiable.
78. 3:34So again, we have this difference between reality and representation,
79. 3:38where they're actually constantly influencing one another.
80. 3:42But it's incredibly dangerous to ignore that distinction,
81. 3:45because this goes beyond our language and consumptive behaviors.
82. 3:48This affects the world we see as possible.
83. 3:51Evidence suggests that ChatGPT is more conservative
84. 3:53when speaking the Farsi language,
85. 3:55likely because of the limited training texts in Iran
86. 3:57reflect the more conservative political climate in the region.
87. 4:00Does that mean that Iranian ChatGPT users will think more conservative thoughts?
88. 4:04Elon Musk regularly makes changes to his chatbot Grok
89. 4:07when he doesn't like how it's responding,
90. 4:09and then uses his platform X to artificially amplify his tweets.
91. 4:12Does that mean that the millions of Grok and X users
92. 4:15are subconsciously being trained to align with Musk's ideology?
93. 4:19We need to constantly remember that these aren't neutral tools.
94. 4:24Everything that ends up in your social media feed
95. 4:26or in your chatbot responses
96. 4:27is actually filtered through many layers of what's good for the platform,
97. 4:31what makes money and what conforms to the platform’s incorrect idea
98. 4:34about who you are.
99. 4:35When we ignore this, we view reality through a constant survivorship bias,
100. 4:39which affects our understanding of the world.
101. 4:42After all, if you're talking more like ChatGPT,
102. 4:45you're probably thinking more like ChatGPT as well,
103. 4:49or TikTok or Spotify.
104. 4:51But you can fight this if you constantly ask yourself: Why?
105. 4:54Why am I seeing this?
106. 4:56Why am I saying this?
107. 4:57Why am I thinking this?
108. 4:59And why is the platform rewarding this?
109. 5:01If you don't ask yourself these questions,
110. 5:04their version of reality
111. 5:05is going to become your version of reality.
112. 5:07So stay real.
113. 5:09(Cheers and applause)