Beyond AI Limits: Data to Wisdom
Key Points
- AI has moved from research labs to everyday life, repeatedly surpassing skeptics’ predictions about what it could never achieve.
- Understanding AI’s capabilities starts with clarifying the hierarchy of raw data, contextualized information, interpreted knowledge, and applied wisdom.
- Many historically “hard limits” of AI have already been overcome, though genuine constraints still remain, making it risky to bet against continued AI progress.
- The future hinges on recognizing where AI excels versus where human judgment adds value, enabling us to combine both for optimal outcomes.
Sections
- AI Limits, Knowledge, and Future - The speaker argues that past predictions about AI’s capabilities were wrong, explains the data‑information‑knowledge‑wisdom hierarchy, and examines which alleged AI limits have been surpassed and what challenges remain.
- AI, Language, Humor, and History - The speaker explains that genuine intelligence requires understanding figurative speech and jokes, illustrating progress from the 1965 ELIZA chatbot to IBM’s Watson winning Jeopardy in 2011.
- AI's Emotional Intelligence Progress & Hallucination Challenge - The speaker notes that AI has achieved simulated emotional intelligence, allowing chatbots to recognize moods, yet points out that hallucinations remain a significant unsolved issue.
- Sustainable AI Scaling & Understanding - The speaker highlights the unsustainable energy costs of ever‑larger AI models, urging smarter, right‑sized model choices while acknowledging unresolved questions about AI self‑awareness and true comprehension.
- Purpose Behind AI Advancement - The speaker reflects on the need for clear goals to guide AI agents, acknowledges the rapid growth and unknown future of the field, and advises focusing on possibilities rather than current limitations.
Full Transcript
# Beyond AI Limits: Data to Wisdom **Source:** [https://www.youtube.com/watch?v=rBlCOLfMYfw](https://www.youtube.com/watch?v=rBlCOLfMYfw) **Duration:** 00:19:44 ## Summary - AI has moved from research labs to everyday life, repeatedly surpassing skeptics’ predictions about what it could never achieve. - Understanding AI’s capabilities starts with clarifying the hierarchy of raw data, contextualized information, interpreted knowledge, and applied wisdom. - Many historically “hard limits” of AI have already been overcome, though genuine constraints still remain, making it risky to bet against continued AI progress. - The future hinges on recognizing where AI excels versus where human judgment adds value, enabling us to combine both for optimal outcomes. ## Sections - [00:00:00](https://www.youtube.com/watch?v=rBlCOLfMYfw&t=0s) **AI Limits, Knowledge, and Future** - The speaker argues that past predictions about AI’s capabilities were wrong, explains the data‑information‑knowledge‑wisdom hierarchy, and examines which alleged AI limits have been surpassed and what challenges remain. - [00:05:01](https://www.youtube.com/watch?v=rBlCOLfMYfw&t=301s) **AI, Language, Humor, and History** - The speaker explains that genuine intelligence requires understanding figurative speech and jokes, illustrating progress from the 1965 ELIZA chatbot to IBM’s Watson winning Jeopardy in 2011. - [00:08:39](https://www.youtube.com/watch?v=rBlCOLfMYfw&t=519s) **AI's Emotional Intelligence Progress & Hallucination Challenge** - The speaker notes that AI has achieved simulated emotional intelligence, allowing chatbots to recognize moods, yet points out that hallucinations remain a significant unsolved issue. - [00:12:22](https://www.youtube.com/watch?v=rBlCOLfMYfw&t=742s) **Sustainable AI Scaling & Understanding** - The speaker highlights the unsustainable energy costs of ever‑larger AI models, urging smarter, right‑sized model choices while acknowledging unresolved questions about AI self‑awareness and true comprehension. - [00:18:00](https://www.youtube.com/watch?v=rBlCOLfMYfw&t=1080s) **Purpose Behind AI Advancement** - The speaker reflects on the need for clear goals to guide AI agents, acknowledges the rapid growth and unknown future of the field, and advises focusing on possibilities rather than current limitations. ## Full Transcript
Artificial intelligence is everywhere right now. In your phone, in your car, even writing
emails for you. You may be wondering if there are actually any limits to what AI can do. I've
heard many people over the last few decades confidently assert AI can do certain things,
but it's never going to be able to do, and then you fill in the blank. Guess what most
of those predictions have in common? They were wrong. The past few years have shown exponential
growth in AI capabilities, bringing it from the research lab to everyday life. And it's doing
most of those things that so many thought it never would or even could do. Of course,
many limitations still exist, but my advice would be this. Don't bet against AI, unless of course
you want to be wrong. In this video, we're going to start with a look at what knowledge really is,
how it differs from data and information, and this will help set the context. Then we'll take
a look at what have been considered to be the limits of AI and see which ones of those things
have actually been accomplished and what's still left to do. Then we'll conclude with some ideas
about the role of AI and humans and where each one excels with the hope of learning how to use this
amazing technology to our best advantage. Let's start off with looking at the relationship that
exists among data, information, knowledge, and wisdom. and we'll use this pyramid to spell it
out. So, we'll start with data. Okay, this is just basically raw facts. If I give you data that looks
like this, I say 10 six uh 42 and 8. Okay, that's raw facts. So, what you don't know really what to
do with that, but that's data for you. Okay, now if I add some context to this data, now we have
information. So this is where we sort of processed it a little more and now I'm going to tell you
that this data actually represents the ages of people in a room. So now we have more context.
This has more meaning to us. Now if I take that and say okay but let's apply some interpretation
to the information that we just had. Now we end up with knowledge. Now knowledge tells us yet more.
So for instance in this case we might say okay I've observed that most of the people in this
room are under the age of 21. So now we've done yet more processing with this. And now finally the
last piece of this is applied knowledge. Applied knowledge now gives us wisdom. and wisdom might
look at this all of this information, all of this data, all of this knowledge and say, you know
what, we've got these people in a room. Let's do something like uh do age appropriate games
to keep them occupied. So, uh the 42-year-old probably won't mind too much playing a game that
a 10-year-old and a and an 8-year-old would play, but you know, they they can go along with that
for a little while. So this is an example very trivial example but you can see what I've done
here data information knowledge and wisdom each one of these adds more context more interpretation
and all of these then lead to the ultimate of wisdom. So another way to look at this pyramid
is data. Well, that's a database. For instance, you know, we can store a lot of stuff in there,
but that's all it is, just a collection of raw facts. Information, okay, we have an application
running on a computer. That's now information technology. That's why we call it that. We've
added context to all of that data. Knowledge, this is where AI really starts to come in. Now, we're
adding more interpretation to the information that we've just processed. But here is where we're
still trying to get. And that's wisdom. Back when I was an undergrad riding my dinosaur to class and
studying AI in its earliest days, there were a lot of things that people said, "These are the limits
of AI. Maybe one day we'll have a system that's able to do these, but they won't be anywhere,
maybe even in our lifetimes." For instance, one of the things that was talked about was the ability
to reason. We needed a system. If we really consider it intelligence, then then reasoning
is a part of that. So the ability to figure out and do problem solving, complex problem solving,
uh this was beyond our capability uh certainly in those days. But since then we've come out with a
computer that can play chess. IBM in 1997 came out with a computer called Deep Blue that played
Gary Kasparov, the best chess player in the world, a grandmaster. That's a lot of reasoning. That's
a lot of problem solving. People thought you'd never have a computer that would be able to beat
a grandmaster. Again, that's already happened. So, what seemed to be a limitation wasn't. Another one
that was really difficult for a long time was natural language processing. Uh, human language
has a lot of nuance, a lot of idioms, things where we say things that we don't mean literally.
And sometimes you're supposed to interpret it literally, sometimes it is figurative speech.
Uh for instance, as I've given examples before, if we say it's raining cats and dogs, we know that it
doesn't mean that there are small animals falling out of the sky. That's an idiom. So we have if a
system is going to really be intelligent in the way that we are. It needs to be able to
understand those things. It needs to be able to understand things like humor and understand
when you're cracking a joke and when you're not. Well, sometimes people can't tell that either,
and sometimes it's because it's a bad dad joke. But be that as it may, in general, we're able to
tell the difference between what is humor and what is not. And we've actually made some advancements
here. In 1965, there came about a first the first of what really is the modern chat bots, but this
was not using modern technology called Eliza. And it was able to have conversations with you. Now,
it wasn't very great conversations, but it would ask you questions and and answer questions.
how are you feeling today? How does that make you feel? Uh this kind of thing almost like you feel
like you're talking to one of these very passive psychologists. Uh but IBM advanced this a lot in
2011 when we came out with Watson which played Jeopardy the uh TV game show and was able to win
and beat champions at that because Jeopardy is full of natural language and play on words, puns
and things like that. You can't program all of those into the system and have it know those. It
really has to understand the meanings behind those things in order to do it. And in fact,
as I say, we've already accomplished that. And look at today's modern chat bots. They're able
to understand a lot of this nuance and they're able to take the instructions you give it in
natural language and understand what you mean in a surprising way. In fact, I think that's maybe
one of the most remarkable aspects of generative AI technology is that it's able to do that for
the first time. We feel like a computer really understands us. It's able to infer what we're
asking for. In some cases, even anticipate the next thing that we need, just like a person
would. We consider that to be intelligent. How about creativity? The ability to create.
I remember hearing a lot of people say, you know, computers can't really create information. Well,
they actually do. Uh, we've got where with generative AI, we can create art. We can create
new works of music. And you can say, well, but those are really just mashups of existing. Well,
guess what? When people compose a new song or draw a new picture, we're influenced by the things that
we've heard as well. Listen to all the top musical artists that you know, and they'll tell you, "Oh,
yeah. Here are my musical influences." So, those things all went into the back of their heads and
influenced the way that they create. So, we are creating new things and they are variations on
the old. But that doesn't mean just because a computer did it, it wasn't creative because in
fact it is. They're coming up with new ideas and will continue to do that. We base our learning and
our creativity on certain things that have been done in the past and so does AI. Now,
here's another one. Real time perception. Things like robots. Well, that was the stuff
of science fiction at one point, but we have them today. And you might not think of it as a robot,
but a self-driving car is one of those where it's having to in real time perceive its environment,
see what's going on, anticipate where the next car is going to move, and where it's going to
be at a specific point in time and do all of those calculations in real time, and make real
uh decisions about that. Robots are having to do the same thing in order to navigate around a room.
So all of these things that basically we used to consider to be limits of AI, I'm going to say,
you know what, we've done all of those. Now, let's take a look at some other areas where we've made
progress, but I don't know if we would say, you know, it's sort of mission accomplished yet. And
one of those would be uh the area of you've heard of an IQ, how about an EQ, an emotional
intelligence uh and an index for that? Well, these systems are able to simulate that. And honestly,
I feel like some people are just able to simulate emotional intelligence as well,
but that's a whole other subject. But an EQ in a system, you can see in the modern chat bots
the ability for them to understand your moods and the way that you're expressing yourself.
So there is some level of awareness in terms of the way that you're describing things. I mean,
we have the stories about people who felt an emotional relationship to a chatbot. Well,
some people feel emotional relationship to their shoe, but that's a whole other thing. The fact
that these systems can talk to us and understand at least give the appearance of understanding
moods and things like that is certainly in the area of okay, I it looks like we're doing this
at least in some cases. Now, another area that's a limitation though that we still have is this area
of hallucinations. Hallucinations are a difficult problem. and they're a a byproduct of generative
AI where the system basically confidently asserts something that just isn't true. So it's trying to
predict what the right answer would be and many many times it's right. It's shockingly right.
But when it's wrong, it is shockingly wrong in these cases. Now we've got technologies that
are making hallucinations less and less likely. Uh things like retrieval, augmented generation
uh helps with this where we feed additional information to give more context so that the
model doesn't just use its own imagination to come up with answers. Uh things like mixture of experts
helps as well where we have different models used for different areas. Chaining of models.
Uh so there are things that we can do in order to reduce the hallucination problem and we're
doing that. So this is one of those I wouldn't say uh is a solved problem but we can certainly see
that we're moving into it. So this one's somewhat solved. Okay. So, those are the things that we've
kind of already done or are still working on and maybe be able to see an end in sight. Let's
move those out of the way. And now, let's take a look at the future. In other words, what are the
current limits? What are the problems that we're still having to to work on these days? Well, one
of the limits of AI is a thing called artificial general intelligence. Right now, we see AIs that
are super smart in a specific area, in a specific knowledge area. Now again with some of these chat
bots that we have today, they seem to know a lot about pretty much everything, but they also have
limitations. For instance, they don't do real-time perception. Uh they can't tie their own shoes,
for instance. So artificial general intelligence would be something that was as smart as a person
doing all the things that we consider to be intelligent and at least on par with what a
person would do across all the different domains. That's something that we haven't really fully
achieved in a single system yet. The next level beyond that would be artificial super intelligence
where we have something that is better than humans in every domain and that's the right
now again the stuff of science fiction. Not saying that we won't do it but we haven't really done it
yet. Another problem that's still to be solved is with sustainability. So right now we have systems
that can do amazing stuff but boy do they suck up the gas. They take up all the electricity.
They need lots of cooling. They're very expensive to run. This is not something that's going to be
able to scale if we just keep throwing more and more processors at this situation. Uh that's not
going to work. We're going to end up using all the electricity that's on the planet just in order to
uh to to run some of these queries. So, we're going to have to be able to make better, smarter
decisions with sustainability. Use models that are the right size, not just the biggest model, but
the right size model. In some cases, a small model might be more efficient and do a better job and
might even hallucinate less if we've got the right use case. So, this is still work that we're that
we're doing that is not yet, I would say, a solved problem, but there's a lot of things we can do
about it. Another one that is really the area that is is science fiction today is self-awareness. So,
is a system self-aware? Does it know it exists? Does it have consciousness? Well, I don't really
know the answer to that. This is really not a computer science question. This is a philosophy
question. So, I'm not going to try to deal with that one here because I'm not even sure how the
answer would be. But another thing that gets us back into this area though is understanding. So,
a system can spit out a lot of things, but it actually understand what it's saying. Um,
does it really know what the meaning of the things are? Seems like it's done a lot of that,
but there's always the question of is this really just simulating? Is it simulating thought? Well,
I don't know. I'll tell you there's a lot of people I've talked to and I think they
may be only simulating thought and simulating intelligence. So, again, it's a little hard to to
draw the line clearly, but uh this seems to be an limitation where the AI maybe doesn't understand
the biggest broadest context that we'd like it to understand. Uh judgment. So remember when I
was talking about data, information, knowledge and wisdom. Well, this is that last one. This is
the business of wisdom and judgment. And in this case, is the system able to make good judgments?
Maybe ethical judgments. Can it determine what is right and what's wrong? Again, can people do
that? Some people have a real hard problem with those kind of judgments. So, it's hard for us
to program a system that will if we can't figure out what those rules would be. But we we certainly
know that right now these are limitations that the systems have. How about in terms of judging
something that's just very subjective like the quality of something, maybe music? You know,
what I think is really great music, you may not think. So, you know, you say, "Well, Jeff,
you have no judgment at all." Uh, but I have a different view of that. But these systems are they
able they're able to generate music and they're able to throw away stuff that is just absolute
gibberish but can they tell what is going to be a hit and what's not going to be for instance in the
music area. So there's a lot of of work here in this space so that it's able to do some of those
qualitative judgments as well. How about this one common sense? And I'm going to really put that one
in uh in quotes because air quotes because um I mean is it really all that common? It seems like
again we have limitations with people. So we can't really expect a system to be able to perfectly do
what we consider to be common sense because we all might have a different idea about that. Certainly
there are some things that we know and the systems ought to be able to understand that but today
there are some certainly some limitations to that. How about in terms of goal setting? Well,
some people would say that with today's agentic AI that a system can in fact set its own goals and
go off and accomplish those things. And what I'm going to make a distinction here is that we have
micro goals. These are sort of the small things that we need to do if I give you a a larger task
and the macro goals. So the larger task, this is what needs to be done. This is how I go about
doing it. And right now today's agents are able to do these kind of micro goals, the goals within
the larger objective, but the big goal, why would we do this in the first place? That's maybe still
uh without uh beyond its reach at the moment. And then sensation, how about this? Does this does a
AI system really sense things? Does it understand what's happening? What how things feel? How things
taste? Um that sort of stuff. The things that are of the senses. Well, we're building robots
that are able to certainly see and hear. Can they taste? In some cases, maybe to an extent,
but there's a lot of other things that go into uh these kinds of sensations that we haven't put all
together in one system. And then here's the really big one, I think, and that's deep emotions. Is a
system really able to feel the same way that we do? Is it able to experience joy? Is it able to
experience sadness, loss, uh, accomplishment? Does it really get what all that's about? And again,
I know some people who don't really do all that particularly well. So, so this is one
of the things that is difficult to put into a system and we can simulate it today, but is it
really feeling these kinds of things? So, I would suggest to you these are some of the things that
to one degree or another are limitations with today's AI. Now, what is the role for humans
and for AI? How do we work together? How do we make sure this is a tool that works for us? Well,
people really should be over here doing this kind of stuff. Answering the what question. What is it
we want to do? That's the overall macrolevel goal, the objective and answering the question why.
What's the purpose of this? Is there meaning in what we're doing? What's the ultimate thing that
we're trying to accomplish? And without purpose, all of this is just meaningless work. So people
are still far better at that kind of thing. And we should be the ones controlling this tool that way.
Over here on this side, once we've told the system what needs to be done, AI can many cases with an
agent figure out the how and go off and perform it, actually do it. Agents are able to automate a
lot of things much faster than a person could. and they can do it in an optimized way, but they need
to know what to do in the first place. We need to know why. So, if you look at a history of AI,
it felt like for the longest time we were making very little progress and then all of a sudden it
just took off. And we're at this this inflection point where the developments and where all of this
is going to go, no one really knows. But what I can say this for sure, we can look at a history
of milestones that we've accomplished already. And we can look at lots of future research, things
that still need to be done, which is actually very exciting. If you're someone that enjoys
it and the possibilities of problem solving, then we're going to be able to do a lot more work and
ultimately we're going to get end up with systems that do things we didn't even imagine yet.
So my advice to you if you start looking at the limitations of AI today I would say don't
become preoccupied with those because the people who have and have asserted that AI will never do
this that or the other thing have generally been wrong. My advice to you, don't bet against AI.