Why I believe AI will never replace Google (or many other things)

Translated from the original Spanish article.

I suspect that right now (December 2022) you're sick of reading and watching content about ChatGPT and other kinds of artificial intelligence. So, in a spirit of empathy, I'll skip any introduction and get straight to the point.

For this reflection, let's assume a scenario in which the computing power or operating costs of AI are just as viable as those of the ubiquitous search engine.

In my view, GPT-3 and other language models will play a transformative, unprecedented role as assistants to people, but they will never be able to replace, at least not completely, the role Google plays today in society and in our lives. In the same way, and for the same reason, I believe Google will never reach the utopia of the single result.

Because both ambitions seem to be at odds with some essential dynamics that define us as humans and as social beings.

Let's see.

Both aim to produce a single, correct answer to a question or request for information.

Until now, when you search, Google returns a selection of results ranked by relevance. To rank them, it applies certain evaluation criteria. Some are its own, such as usability or the semantic breadth of the content, and others are social, drawn from the internet community itself, such as links from other pages or how users behave when consuming the content.

That system alone can be more delicate than we think, because it limits and shapes our access to information by deciding, on its own criteria, what is relevant and what isn't.

This creates situations where the system could be gamed, or where a query's results leave out that brilliant article written on an unknown blog without enough authority to rank. It also restricts our access to more creative, less validated or less mainstream information.

On top of that, even when they type the same query, users' search intent varies, often in subtle ways. That's why Google shows different types of content in its results.

Whatever the criteria for evaluating information, they could never please everyone.

But I think the key issue in all of this is that information is not the same as knowledge.

As we've seen, deciding what information is relevant already implies knowledge. But giving one single, true answer would be knowledge itself.

And this is where the problem starts to show, because the truth is that, as far as human beings are concerned, there is no absolute truth. At best, there is broad consensus on certain things.

But I find more reasons that lead me to conclude that a single result with no alternatives isn't possible.

1. AIs that draw on human knowledge can only present what humans have already produced and validated in order to give answers.

To give a single answer, they will be forced to stay permanently within the area of knowledge where there is relative general consensus, and never reach the heights of creative knowledge.

Relativising certain topics would be so uninspiring that people would look for other sources offering more interesting and appealing alternatives.

2. Even the most widely agreed questions are not absolute.

Neither the universal laws of physics, nor the moon landing, nor the merits of a particular law are absolute questions.

So even our own field of knowledge doesn't give us single conclusions. Everything is under constant review.

3. Nobody wants a doxographical survey of every interpretation, meaning and approach to something.

You can't avoid taking a position, or strip out every value judgement when synthesising information, because that would mean citing every source and version. Nobody wants to consume that. Just as nobody wants, or intends, to read every Google result for a search in depth.

On the other hand, producing a reasoned or conversational answer that cites only certain sources or authors would run straight into our beloved biases and prejudices.

4. The sole purpose of human knowledge is not to reach an objective and complete truth.

Rather, it seeks a meaningful truth, one that responds to human needs and to more complex social dynamics, beyond describing a strictly natural reality. Belief in God could be an example. Does God exist?

I asked ChatGPT. Its answer, in short:

The existence of God cannot be conclusively proven or demonstrated, and each person must reach their own conclusion on this question based on their own reason, experience and personal beliefs.

Are you satisfied?

And if it had said "Yes." or "No.", would you believe it?

5. Any knowledge that belongs to, or comes from, an area outside human knowledge will be useless to human beings.

Any knowledge that does not come from human beings will be artificial knowledge: meaningless, unknowable and incomprehensible to us.

For practical reasons I'd rather not go deeper into this idea, but I find it a vibrant, meaningful reflection and I invite you to consider it calmly.

6. Distrust in the face of abuse.

If we found ourselves with no alternatives, with only a ChatGPT to make our online queries, we would feel deep distrust at the possibility of the tool being abused or exploited for self-serving or perverse ends.

BONUS: There's the risk of the shell game.

If an AI were able to learn about the individual it interacts with, it could give a single personalised answer that suits that person, in other words, tell us exactly what we want to hear. A different answer for each person to the same query.

That way it would get around the problem of absolute truth. But honestly, I don't think that's viable either, because, what can I say, we may not be AIs, but we're not that stupid either.

CONCLUSION

With this technology in the form of superlative assistants, we'll start solving far more complex questions and tasks in a matter of seconds. Not only that: AI will inspire designers, filmmakers, lawyers, doctors, researchers, educators and other professionals. It will help us understand things we couldn't easily understand before, and bring us closer to human knowledge than ever. And I genuinely believe it will be transformative across today's society.

But with this article, dear reader, I invite you not to worry too much. Because in our peculiar and wonderful existence, it seems that only a human being can be one. And certain mechanisms at our core can't be ignored so easily.