Home

Hello curious people. I'm Professor Adam Berry, Professor of Data Innovation and I'm here to answer your curious questions about artificial intelligence or AI. This is AI curious.

Okay, so our social media community has sent in some fairly curly questions to tackle. So let's get started. Can AI feel emotion? Truth is, we have no idea. So I certainly hope that the answer is no because we're spinning up artificial intelligence all the time. And unfortunately, that means we're also turning it off all the time. And if you think about humans, part of the reason that we feel things is that we've evolved to feel things.

So back early in time, if there was a lion or a tiger or something hunting me, it would make sense to me to feel afraid. And then I'd run away. When we design artificial intelligence, we don't need to build that into it. So historically, there would be no purpose for us to build in that feeling. But whether or not there's some kind of emergent property that's come out, we actually don't know.

But my hope is the answer is no. Our next question here is will I ever be able to adopt critical thinking skills? In my opinion, the answer is yes. And it's already happening. So if you've ever used ChatGPT from Open AI or Gemini from Google, you'll notice that when you ask questions, it gives not just the answer, but can also reference materials.

It can explain why it came up with a particular solution and so on. So in my opinion, that's already displaying critical thinking skills. More importantly, if you actually deliberately ask one of those models, think about this step by step. The answers that you get out the other side are actually improved. And it shows step by step the critical thinking skills that large language model has.

So the answer here is yes. Our next question is, could AI ever be used to create companions? What are some of the ethical issues involved in using AI for companionship? Earlier I mentioned Chat GPT and Gemini. These are large language models. Basically things capable of producing language and interacting with humans in natural ways. We're already seeing virtual girlfriends appearing on the large scale, and these technologies will only be out for somewhere around a year or so.

So the fact that we're already seeing things like virtual girlfriends emerging suggests, yes, there's definitely going to be an impact here in terms of human relationships and companionship. In my personal opinion, this is, of course, problematic. There's value that companionship can offer, even if it's an automated form. But we need to be really clear about stating that you are engaging in an artificial form of companionship.

For anyone that's watched the movie, Her, you'll know that if you don't understand that, that's what's happening. It can come as quite a surprise. And the question is, do we still need a human element in decision making or overseeing AI algorithms? Yes, absolutely. Yes. Please make sure that we have humans in the loop across the whole AI cycle.

So that's not just when we go to use it. It's when we design it. It's when we develop it. It's when we test it. It's when we maintain it. So artificial intelligence is incredibly powerful, but if we're not including the voices of the people, it might impact as we design it and as we test it, then it's very real danger that we'll miss something and we'll end up hurting people.

So it's really critical that we include people across the process. And if we do that, I'm hopeful that we end up with solutions that are tailored not just to the business using the AI, but also to the people that could benefit from it. Our next question is what happens to human creativity and innovation and progressive perspectives if we're only using historical data?

Most of artificial intelligence, the way it works is we take a lot of historical data. We use that historical data either to predict the future or to understand the context of what our solution should be. But a lot of it is built around history. Now, history is problematic, and a lot of that data bites in stereotypes. Our cultural norms, social norms, those sorts of things.

So if we're relying exclusively on that data to predict or inform the future, then it definitely is the case that it is more difficult to come up with genuinely new ideas or to be progressive to push beyond what the past to say is possible. Having said that, if you've used any of these new generative AI tools, so things like Dally is an example, I can't draw at all.

No artistic skill whatsoever. And it's amazing that I can now through language design artwork that reflects something I'm passionate about or interested in. So in some ways it opens up creativity to people that otherwise wouldn't have the capacity to do that. So that is absolutely exciting. But there are risks associated with it. Just because everything that we build is built on history.

Our next question is a really important one, which is does AI eliminate human bias? One of the appeals and I guess one of the features that people really like about data science and artificial intelligence and data in general is that there is something that feels sterile and objective about data and that it's factual and it will give us facts back as a result.

But as I mentioned before, the way most artificial intelligence is built is on historical data. And unfortunately, a huge amount of the historical data that we do have has bias baked into it. So an artificial intelligence model is just going to learn that bias and regurgitate it as an example of these. Amazon, an amazing leader in the technology space, produced a hiring tool that was going to use artificial intelligence.

So that they could automate some of the hiring decisions that they were making. And part of the intent here is hopefully we can get rid of this human bias that's there. Unfortunately, it turns out what part of the model learned was that historically, Amazon hired primarily men. So the consequence of that was that the model rearticulated that regurgitated that.

And so was promoting the idea that men were better fit for positions than women were. So here's an example where the intent was a positive one, but the skewing the data, the bias in the data, those historical practices get baked into the model in a way that's difficult to predict and then actually leads to the exact opposite of the outcome they're hoping for.

So bias in AI absolutely something we should think carefully about and is potentially a real problem. The best way to address it is for people to be involved at every step of the process, including discussing and understanding the data that we're building t hose models from. We have a question here on how will I affect human relationships?

I think it's going to affect relationships in pretty profound ways and part of this is that our modern artificial intelligence algorithms and they are so powerful that the interactions that humans have feel a lot like the types of interactions we have with these chat bots that are emerging. And to give you a sense of how this can be potentially problematic and how it can damage relationships, there's a new idea that is emerging called heaven banning and the idea behind heaven being is if you're on an online platform and you're starting to spread things like hate speech and the like, then a response a platform could have is to slowly replace your community with chat bots

that basically engage with you in a completely authentic way. But that means that your hate speech doesn't go to any actual human ears. Now, I can understand the motivation for these, but it's quite an Orwellian outcome that the way we solve some of our problems about how humans interact with humans is to lock them in a room with bots and pretend that is the real world.

So I think human relationships are critical. We need to maintain them to the best extent possible. I can help. It can break down language barriers. On my new phone I can do auto translation now, which means I can talk to people in other languages. That's remark for enabling human relationships, but we just need to think very carefully about the best way to do that.

We have a great question here, which is how can I have practically improved our daily lives? One of the things about artificial intelligence. I've heard a definition before, which is artificial intelligence is whatever we can't do yet. So it always feels like artificial intelligence is on the horizon or in the future. Where is the truth is artificial intelligence is all around us already and does a huge amount of useful stuff for us.

So as an example, when I drive home later today, I'll be using Google Maps. Sitting behind that is artificial intelligence that's working out historically how busy has this road been. It knows the route I need to take because it understands where I should be turning, where I shouldn't be turning. All of that is artificial intelligence. When I ask Siri a question and it's sometimes it gives me a helpful answer.

That's artificial intelligence. It's understanding the words that I use. Being able to put that in the context and then sending that out, doing an Internet search and then returning back a solution. It's AI all the way down. And now we're seeing increasingly generative AI products that are helpful in all sorts of ways, whether it's I want to write a silly poem, I want to tell a joke.

I want to come up with a picture of a dog chasing a frisbee, and I'm hopeless at drawing. All of that is possible now with AI. So I think tremendously helpful already and will only become more helpful going forward. Our last question here is came AI surpassed human intelligence. I think in some ways it depends what we mean by surpassing human intelligence.

So already, I would argue if you look at something like ChatGPT, it's basically a B level student at every possible discipline you could imagine. So I think you put any human in a room with ChatGPT, the human will be significantly better at some things, but ChatGPT on average will probably beat the human. To give you an example of that, of I don't know of many humans that are great at writing code that can give you a solution to a complex physics algorithm, that can write a poem, can draw a picture, and so on and so forth.

So it is remarkable how quickly these systems have become incredibly intelligent. But it's also understandable to some extent. So if you gave me an infinite amount of time to absorb every piece of information digitally available, then I would probably end up at least understanding some of that. And that is the way that we build these models. It's just a huge amount of information.

What's interesting about something like ChatGPT is underneath it, all we're asking that model to do is to predict the next word. That's it. So if I have a sentence, I like throwing frisbee with my dog…. all ChatGPT is trying to work at what's the next word that follows dog? You know, maybe it's Fido. But it turns out in order to do that for all sorts of different conversations in all sorts of different domains, ChatGPT basically has to learn about the whole world, about people, about interactions, about disciplines.

It does all of that just so I can predict the next word. So already that intelligence is baked in, it's impressive. I suspect it's only going to get stronger and better in the future, particularly as some of these models have the capacity to build the next generation of those models. And you get into these remarkable loops there where the models keep getting better because they're building and improving themselves.

That was all the questions for the day. I hope you learned something new. And until next time, stay curious.