Managing AI - CompTIA A+ 220-1202 - 4.10 — Transcript
Full transcript
- 0:02AI is an abbreviation for artificial
- 0:04intelligence and the goal of AI is to
- 0:07meet or exceed human intelligence in the
- 0:10form of technology. This means that our
- 0:13computers would be able to learn, infer,
- 0:16and reason based on input that we
- 0:18provide. This is obviously not a new
- 0:21idea. We've been working on concepts of
- 0:23AI for a number of decades. And of
- 0:25course, you see AI mentioned in numerous
- 0:27science fiction stories. But now, of
- 0:30course, AI has become much more
- 0:32prevalent. We have new tools that allow
- 0:33us to create new content such as text,
- 0:36audio, and video by simply asking a
- 0:39computer to create it.
- 0:42And very quickly, we have seen AI become
- 0:45integrated into most applications that
- 0:47we use every day. For example, if you
- 0:49perform a search on a popular search
- 0:51engine, you may find not only your
- 0:53search results, but also a summary of
- 0:56those results that was created using
- 0:58artificial intelligence. The goal is to
- 1:00take information that normally you would
- 1:02need to find across multiple sites and
- 1:05present all of that information to you
- 1:06as one single AI summary. We've also
- 1:10seen AI in our email clients and on our
- 1:12mobile devices. Instead of having to
- 1:15read through a long email, you can read
- 1:17an AI summary that explains the
- 1:19information that was contained in that
- 1:21message. And as you've probably seen in
- 1:23graphics editors, you can use generative
- 1:25AI to fill in or remove information from
- 1:28an image by simply highlighting it on
- 1:30the screen and asking the AI to remove
- 1:33it. We've also seen artificial
- 1:35intelligence engines that allow you to
- 1:37input a description of what you'd like
- 1:39an image to be and the AI will create
- 1:42that image on the screen.
- 1:45We've seen a number of different uses of
- 1:47AI that we could put into the category
- 1:49of appropriate AI use. For example,
- 1:53there may be extremely large data
- 1:55repositories that we would like AI to be
- 1:57able to sift through and provide us with
- 2:00a summary of that information. For
- 2:02example, there's AI that can look
- 2:03through terabytes and terabytes of log
- 2:06files in an organization and find
- 2:08potential security issues that may be
- 2:10hidden in all of these different logs.
- 2:13This would certainly be timeconuming and
- 2:15difficult for a human to do, which makes
- 2:17it perfect for AI. We've also started
- 2:20integrating AI into our scripting and
- 2:22our automation so that AI can make
- 2:24decisions on what to do when a certain
- 2:26situation occurs. Being able to do this
- 2:29without human intervention means that we
- 2:31can spend our time doing something else.
- 2:34We're also seeing AI integrated more
- 2:36into health care. For example, AI can
- 2:38take the output from an MRI scanner and
- 2:41evaluate that to see if there may be
- 2:43something that would be concerning. We
- 2:45might also have AI go through healthcare
- 2:47records and determine if anyone is
- 2:49receiving multiple drugs that might have
- 2:51negative interactions. This means the
- 2:54health care professionals can work with
- 2:55us at a human level and have AI take
- 2:58care of everything at a digital level.
- 3:01And if you're watching this video on
- 3:02YouTube, you may be getting realtime
- 3:04language translation thanks to the AI
- 3:06that's built into our new communications
- 3:08apps.
- 3:10But of course, there are downsides to
- 3:12any technology and AI has been used for
- 3:15things that may not be as appropriate.
- 3:17For example, we've seen AI used for
- 3:19fraud. AI can impersonate a real person
- 3:22and even create video of that person,
- 3:25something we refer to as a deep fake.
- 3:28We've also seen application code that's
- 3:30created with AI without any knowledge of
- 3:32what the application is actually doing.
- 3:34Or perhaps someone is using AI to create
- 3:37an image and taking credit for that
- 3:39image creation. And we've seen AI used
- 3:42in schools and colleges to plagiarize
- 3:44works. Students are using AI to
- 3:46summarize existing works, but they're
- 3:48not citing AI in that summary.
- 3:52One of the challenges we have when
- 3:53working with AI is that it can sometimes
- 3:55have a bias. The AI only knows the
- 3:58information that we're feeding into it
- 4:00and it has a predefined set of
- 4:02algorithms on how to manage that data.
- 4:04And sometimes AI can make the wrong
- 4:07conclusions based on this combination of
- 4:09data and algorithms. For example, we may
- 4:12be feeding AI a very large data set
- 4:15associated with health care in the
- 4:16United States, but inside of that data
- 4:19set, we may have under represented
- 4:21statistics for a particular ethnic group
- 4:23or particular gender group. In this
- 4:25case, AI could potentially create
- 4:27conclusions that are biased one way or
- 4:30the other. We're obviously not trying to
- 4:32create bias as part of this AI analysis.
- 4:35So we have to make sure that we create
- 4:37an AI algorithm that doesn't build bias
- 4:40when we have these large data sets. A
- 4:42good example of this is what Amazon
- 4:44found when they took 10 years of
- 4:46submitted résumés and put them into an
- 4:49AI engine. This grouping of résumés had
- 4:52a number of terms such as executed and
- 4:55captured as part of the résé text. The
- 4:57artificial intelligence engine tended to
- 5:00pick résumés that had those terms as
- 5:02part of the resume itself. What we found
- 5:05is that those terms were more often used
- 5:08when it was a male submitting a resume
- 5:10versus a female and therefore the AI
- 5:13created a bias towards male résumés.
- 5:16Obviously, Amazon chose not to use this
- 5:19AI algorithm for their resume research.
- 5:23There's also occasions where AI can
- 5:25completely misrepresent the data that it
- 5:27was provided. You can think of this as a
- 5:30confidently incorrect AI where it really
- 5:33knows the answer but the answer is
- 5:35completely wrong. For example, let's say
- 5:37that we've told the AI that any picture
- 5:39containing a snout, a tail, and four
- 5:42legs is a dog. Then we provide the AI
- 5:45with a picture of a cloud and the cloud
- 5:47appears to have a nose, a tail, and four
- 5:50legs. Therefore, the cloud is obviously
- 5:52a real dog. This is obviously a
- 5:55simplified way to describe an AI
- 5:57hallucination, but this is what tends to
- 5:59occur when we have this type of data
- 6:01misinterpretation. These are also not
- 6:04very unusual to find, and you may have
- 6:06seen an AI hallucination when performing
- 6:08your own tests with an AI engine. A
- 6:11practical example of an AI hallucination
- 6:14was when Google and Microsoft announced
- 6:16their new AI technologies. Google
- 6:18announced Google Bard. This is now
- 6:20Google Gemini. and Microsoft introduced
- 6:23Microsoft Bing Chat. Both of these
- 6:25engines provided a summary of some
- 6:27articles that had completely wrong
- 6:29information inside of them.
- 6:31Unfortunately, this hallucination was in
- 6:34the announcement that Google made
- 6:36regarding their new AI engine. This AI
- 6:39engine took an input that said, "What
- 6:41new discoveries from the James Web Space
- 6:43Telescope can I tell my 9-year-old
- 6:45about?" And then it gave a number of
- 6:47bullets that you could use to help
- 6:48describe the James Webb Space Telescope
- 6:51and its findings. Unfortunately, one of
- 6:53these findings was that the James Webb
- 6:55Space Telescope took the very first
- 6:57picture of a planet outside of our own
- 7:00solar system. Bruce Macintosh replied to
- 7:03this and said, "Speaking as someone who
- 7:04imaged an exoplanet 14 years before the
- 7:07JW Space Telescope was launched, it
- 7:10feels like you should find a better
- 7:12example." This points out that even the
- 7:14best AI engines can suffer from an AI
- 7:17hallucination.
- 7:20Behind the scenes, artificial
- 7:21intelligence is creating a conclusion
- 7:23using a model that is provided. And
- 7:26sometimes it makes the wrong conclusion
- 7:28based on the model that it was given.
- 7:30And researchers will often perform tests
- 7:32of an AI engine to see how accurate it
- 7:35might be. They will get predictions from
- 7:37the AI and then they will compare those
- 7:39predictions to known test data. One of
- 7:42these accuracy tests was done by
- 7:44originality.ai. They did an AI
- 7:46fact-checking accuracy study in August
- 7:48of 2024, and you can read through the
- 7:51study yourself at the URL that you see
- 7:53here. They evaluated their own AI and a
- 7:56number of other AIs that were currently
- 7:58available. The originality.ai
- 8:00engine got a 72.3%
- 8:03accuracy. GPT4 got a 64.9%
- 8:07accuracy rating and GPT3.5 got a 58.6.
- 8:11six accuracy rating. As you can see, our
- 8:14current iterations of AI are far from
- 8:16perfect and in some cases fall well
- 8:19under what most people think AI is able
- 8:21to do.
- 8:23There's been a lot of work made on both
- 8:26public AIs and private AIs. A public AI
- 8:30would be one that is publicly available
- 8:31on the internet. Things like ChatGpt or
- 8:34Google Gemini are good examples of a
- 8:37public AI. But some companies are
- 8:39creating their own AI engines and these
- 8:41would be engines that are private to
- 8:43that organization. These would commonly
- 8:45have proprietary company data and the
- 8:48organization has complete control over
- 8:50the modeling used for that artificial
- 8:52intelligence engine. The decision on
- 8:55whether you use a public AI or a private
- 8:57AI may be based on data security.
- 9:00Whenever you put information into an AI
- 9:02engine, there is a possibility that that
- 9:04data could be retrieved. And if that
- 9:06data is sensitive or contains private
- 9:09information, you could be putting that
- 9:11information available for anyone on the
- 9:13internet to see. We've also seen cases
- 9:15where an AI was able to provide
- 9:17passwords, encryption keys, certificate
- 9:19details, and other sensitive
- 9:21information. Of course, if you're
- 9:23running a private AI, only people within
- 9:26that organization have access to that
- 9:28proprietary data, and therefore, it
- 9:30limits the scope of any type of breach.
- 9:32There's also quite a big difference
- 9:34between a public AI and a private AI
- 9:37when it comes to how much data they're
- 9:38able to evaluate. Obviously, a private
- 9:41AI is only going to have data from a
- 9:43single entity, but a public AI is
- 9:46collecting data from everybody who's on
- 9:48the internet. In those cases, it may be
- 9:50that your public AI is more accurate
- 9:52than the private AI because the public
- 9:54AI has access to so much more data. But
- 9:57it is that large amount of data that
- 9:59does create concerns about privacy. Some
- 10:02of that information that's being stored
- 10:04by a public AI is information about you.
- 10:07The AI knows where you live. It might
- 10:09even know your previous addresses, has
- 10:11information about your habits and things
- 10:13that you do on the internet, and it may
- 10:15have details about any memberships and
- 10:17groups that you belong to. Let's say
- 10:19you're applying for a job. What type of
- 10:21summary would the AI create of you and
- 10:24what would be contained within your
- 10:26profile?
About this transcript
This page contains the full transcript of Managing AI - CompTIA A+ 220-1202 - 4.10 by Professor Messer, generated from the public captions YouTube serves with the video. The transcript has 1,775 words across 266 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.