Read the full transcript of John Tasioulas’s talk titled “Why AI Is A Threat – And How To Use It For Good” at TEDxAthens 2025 conference on Feb 16, 2025.
Listen to the audio version here:
TRANSCRIPT:
JOHN TASIOULAS: Well, I’m a philosopher in Athens in 2024, talking to you about artificial intelligence. That is a great thrill and a great honor. But you might ask, what does the ancient discipline of philosophy have to say about the AI revolution?
I’ve got 15 minutes to make the case. Let’s go back to Socrates in the Republic. He says, the question we’re dealing with is not a trivial question. It is the question, how should one live? Now this question, how should we live, is a central philosophical question.
But it’s also a question each one of us has a responsibility to answer for themselves. That’s why Socrates conducted his dialogues in the Agora in Athens with ordinary citizens. And AI makes it all the more urgent to ask this question again, because this technology is so revolutionary, it could transform our lives, both for the better and for the good. So we need to ask this question again, in light of these new technological developments that Socrates could not have foreseen.
But not only does AI make this question, how should we live, more urgent, it creates problems for us in addressing this question, new problems, new threats to our ability to answer Socrates’ question. So today I’m going to talk about three of these threats.
Threat 1: Distortion of Self-Understanding
The first threat is that AI threatens to distort our self-understanding, our understanding of what it is to be human. All the big tech corporations say they have the same goal. Their goal is to create artificial general intelligence. Artificial general intelligence is intelligence that is replicating the entire spectrum of human intelligence. So able to write a poem, make a cancer diagnosis, make a hiring decision, do everything that a human can do.
But if we pursue AGI as our aim, there will be a temptation, an incentive, to imagine that AGI has been achieved because we’ve changed the goalpost, we’ve changed our own understanding of what humans are like, blurring the distinction between humans and machines. But we need to preserve this distinction.
So what are the differences? Well, one fundamental difference relates to understanding. An AI system like GPT-3, large language model, operates on the basis of statistical correlations between data. These statistical correlations do not give it a genuine understanding as a human would have of what a cat is or what an electron is. Humans do have that understanding, partly because we’re embodied agents engaging with the physical reality. And one way this difference shows up is that there is a quality that most humans have, which is common sense, which has proved very elusive for AI systems. In other words, AI systems make spectacular mistakes that no human would ever make.
They confuse a cat with a skateboard or a human with a gorilla. No human would ever make these mistakes because humans have a rooted understanding of the world in a way that a machine operating on correlations between data points does not. So understanding is a big difference.
But there’s another big difference, and that is humans have a capacity for rational autonomy. We have the capacity to choose our goals, to decide, do we want to be a doctor or an actress?
Do we want to be a lawyer or a musician? And to choose these goals in light of all the reasons, pro and con, for each option. An AI system cannot do this. It has a goal programmed into it, and it optimizes for the fulfillment of that goal. Now you might say, if we start to blur the distinction between human and machine, what difference does it make?
Why is this important? Why is it a threat? The reason it’s a threat, I think, is the reason given by Aristotle, which is if we want to know what a good life for a human being is, we need to know what the distinctive capabilities of a human are. These capabilities for reasoning, for communicating, for social engagement that manifest themselves in all sorts of valuable pursuits, like scientific research, or sustaining a friendship, or fighting for justice. If we lose our vivid sense of the capabilities we have as humans, the risk is we will have an impoverished ethics. We will have an impoverished sense of the values that matter.
What will that impoverished ethics look like? Well, one form it will take will be a kind of consumerism that says what matters is your gratification as a consumer, which can be just a passive matter of receiving pleasure rather than exercising your capabilities. Or there’s another kind of ethics that’s become very popular in Silicon Valley, which says the pathway to a better life is transhumanism, is to transcend our human nature, to live in virtual reality, to become cyborgs. Now in the Aristotelian view, according to which a fulfilling life is exercising your human capacities, transhumanism, which is taken seriously in these powerful circles, is not a path to utopia, it is a path to species suicide.
Threat 2: Distortion of Problem Understanding
Another threat, astronomical sums of money are being invested by the private sector in the development of artificial intelligence. In 2025, it is projected that $200 billion will be invested in developing AI technologies. And there will be a profound economic incentive for these companies to say to us that these AI systems can solve problems better than humans can. And that will create another distortion. It will create a distortion whereby increasingly, maybe unconsciously, we start to change the nature of our problems, the way we understand our problems, to make them more suited to being handled by AI systems. So let me give you an example of this.
Risk assessment tools based on AI technology are already used in criminal justice.
So one question in criminal justice is the question of bail.
Should someone who has been accused of a crime be allowed to go free, awaiting their trial, or should they be kept in custody? Now there are a lot of very prominent thinkers who say AI systems are better at this kind of decision than human judges.
They say the argument is in two parts. One part is the bail decision is fundamentally a prediction. Will this person commit an offence if they are released?
And the second claim is AI systems are free from certain biases that humans have and they are better at making these predictions. Now that second step in the argument is very controversial.
It’s not entirely clear that AI systems, far from it, are better at making the prediction. But I want to challenge the first step of the argument, the way the problem is understood.
Is the bail decision exclusively a decision about will this person commit an offence? Clearly not. There are other considerations that are relevant, like how serious was the offence that they committed?
What will be the impact on their children and family if they’re kept in custody? What’s the capacity of the prison system to house this person?
So when we think about this bail decision, it’s not a monodimensional decision based on a prediction. It requires balancing a whole series of factors. And there may not be one single correct way to balance all those factors. There’s not a way where you could say, I’m going to give all these factors a numerical score and then choose the option that gets the highest number. That’s not the way the situation works, but that is how an AI system would work. So the worry here is, instead of using AI to address our problems, we change the nature of our problems.
That AI starts dictating to us, as it were, what the problems are that we’re confronting. We’re getting things back to front.
Threat 3: Threat to Our Values
A third threat I wanted to talk about is to our values themselves. People who now push AI systems, especially as replacing human effort, tend to focus on two things. One, that these AI systems will produce valuable outcomes. They’ll produce a correct cancer diagnosis, a good hiring decision, or good journalistic copy. And second, they say, they’ll produce these good outcomes efficiently, faster, and at less cost than a human. But this relentless focus on valuable outcomes produced efficiently ignores a whole range of values that we care about.
Values that I’m going to call process values. Values about the way in which we produce valuable outcomes. And I think this is basically Cavafy’s wisdom in the poem Ithaca. That what matters to us is not just reaching the final destination by the quickest route. We also care about reaching that destination in the right way. How we reach that destination. Arriving in Ithaca in my private helicopter is very different from arriving there by a hero’s journey, encountering man-eating giants, Cyclops, and angry Poseidon. And sometimes the journey is more important than the destination. Remember the line, “and if you find her poor, Ithaca won’t have fooled you.”
Now, how does this work out in real life? How is it that we care about process and not just outcome?
Well, think about loving relationships. Yes, we seek loving relationships. But it makes all the difference if we’re in the relationship through a free choice that expresses our personality and our tastes.
And also, we assume the risk of making a terrible mistake. That’s part of the value. As opposed to being assigned to a relationship by an algorithm that has been configured to optimize for human compatibility. Or think about work. At work, we want to produce valuable outcomes, valuable goods and services. We want to have the outcome of a decent salary. But we also care about how these things are produced.
We want to be able to work in a way that involves the exercise of skill, the exercise of judgment, the possibility of real cooperation. These are process values and they will never be fully compensated by the idea that there will be a universal basic income to compensate people for jobs lost through automation. Or take a final example, the case of the judge. We want good legal decisions that are correct. But we also want decisions that are made for the right reasons.
There’s an AI system called Lex Machina, not Ex Machina. And Lex Machina can predict the outcome of legal cases in US patent law just as well as the top US patent attorney. Sounds impressive. But then I’ll say to you, the way it does this is not by looking at the law and making a prediction.
It looks at the name of the judge. It looks at the amount of money at stake in the case. It looks at the law firms involved in the case. These are irrelevant to what the correct outcome should be. So you get correct decision, but wrong reasoning.
It’s like the AI system that was able to distinguish pictures of huskies, certain kind of dog, from wolves. And it was very successful. But it was using the presence of snow in the picture to make the distinction.
Right outcome, wrong reasoning. We also want a judge who, if they’re going to make a decision that affects our life and liberty, can take personal responsibility for that decision. AI systems lacking rational autonomy cannot be held responsible in the same way. And this is very important because one of the things we need judges for is for the dreadful job of taking responsibility for decisions that have a huge impact on human life. But it’s also a human tendency to try to avoid responsibility where possible.
AI systems are going to be a new way in which we can shirk our responsibilities for the terrible consequences we visit on our fellow citizens. And finally, there’s the point about empathy. Even if the AI judge is merciful to me and gives me a reduced sentence because of the hardship that I was facing, that led me to commit the crime, that same sentence will not have the same meaning as if it were delivered by a human judge. Because the human judge can deliver that sentence expressing a compassionate reaction to the situation I confronted. He could say, there but for the grace of God go I. The AI system cannot say that.
It doesn’t share the human life with us. I’ve talked about risks AI poses. But you might now ask, what are the solutions?
Solutions and the Role of Democracy
And I think on any realistic view, there is no simple single solution. For example, in my institute, we do work on things like the need for a new right, the right to a human decision, a special new right for the age of AI. But if there is one more fundamental solution, I think the answer lies in democracy.
Because the historical record shows that technical and technological innovation does not automatically bring benefits for everyone. It only brings prosperity and other benefits if it’s subject to robust democratic control. Democracies are better at producing good decisions than non-democratic systems. And also democracies have the kind of process value I talked about. If we deliberate together as free and equal citizens about the common good and reach a decision together, we respect each other’s dignity in a way that other systems don’t. OK, but of course we know that we live at a time of democratic crisis, of loss of faith of democracy.
We live at a time when there’s a rise of technocracy, more and more decisions being taken away from democratic publics and given to experts, judges, bankers. We live in a time of populism where people look for a strong man ruler who can act outside of democratic structures. And insofar as AI is involved in all this, a lot of people think AI is toxic for democracy because it enables the spread of misinformation, disinformation. It foments political polarization. I want to end with a more positive story. And that is, although that is true, there’s also another vision that’s possible for AI, which is to enable a more participatory form of democracy that is also better informed.
Now, the obvious objection that people make to participatory democracy is they say that was fine for ancient Athens, but it’s not fine for modern states given their size and their pluralistic character. But I think the real top of the agenda item for us now should be using AI and digital technology to enable this more radical participatory democracy that we need.
So you could imagine AI tools providing citizens with information that is tailored to their specific learning style. You could imagine them bringing together random samples of affected populations and promoting and moderating deliberation and debate. And they could circulate rival proposals and identify points of consensus, not just simple majority, but points of consensus that are across different democratic groups. Now, you might say that sounds like a philosopher’s dream, but in fact, it’s already in existence.
In Taiwan, for example, there was a grassroots movement that became the government. And this grassroots movement used an online platform called Polis. I wonder where they got that name from. Polis enables citizens to come online to engage in deliberation and debate about questions like, what should our policy be towards Uber?
And then their deliberations actually feed in to final legislation. So I think we’re living at a time in which people feel increasingly disempowered, subject to forces they can’t control or even understand.
Conclusion
And I think there’s a serious threat that AI will exacerbate that condition of disempowerment by creating a dehumanised world in which the exercise of our distinctive human capacities is pushed to the margins. And all of this will be done under the happy banner that it’s satisfying our consumer preferences. And there will be powerful economic interests behind all this. It’s not going to be easy to fight back, but the fight back has to begin by thinking about what our deepest values are that are at stake here. The fight back has to begin by going back to Socrates’ question. Thank you.
