Artificial Intelligence and it's impact on the world.

Artificial intelligence is expected to revolutionise your life, and possibly the entire globe, in the coming years. However, they can't seem to agree on how.

The following are excerpts from an interview with famous computer science professor and AI expert Stuart Russell, in which he helps separate the sense from the nonsense.

An Interview with Stuart Russell

There is a significant difference between asking a human to accomplish something and assigning that task to an AI system.

When you ask a human to get you a cup of coffee, you are not implying that this is their life's work and that nothing else in the universe matters. Even if it means killing everyone else in Starbucks to get you your coffee before it closes, they should do it.

That is not what you mean.

All of the other things that we both care about should be reflected in your behaviour as well.

The problem with AI

And the problem with how we currently develop AI systems is that we give them a fixed goal. The algorithms need us to specify every aspect of the goal.

And, if you ask, can we stop ocean acidification? You could have a catalytic reaction that achieves that really efficiently, but it burns a quarter of the oxygen in the atmosphere, causing us to die slowly and inconveniently over several hours.

So, how can we avoid this issue?

You can argue, well, just be more specific about the goal—don't forget about the atmospheric oxygen. Then, of course, a byproduct of the process in the water poisons all the fish. Okay, I meant don't kill the fish as well. Then there's the seaweed, of course. Don't do anything that will kill all of the seaweed. And so on and so on.

Humans have no idea what they care about

And the reason we don't have to do that with humans is that humans frequently recognise that they don't know all that we care about.

If you ask a human to get you a cup of coffee and you happen to be in the Hotel George Sand in Paris, where the coffee costs 13 euros a cup, it's quite reasonable to answer, well, it's 13 euros, are you sure you want it, or could I go next door and get one?

And it's entirely normal for a person to do so.

I'm going to repaint your property; is it acceptable if I remove the drainpipes and then replace them?

AI systems lack it

We don't consider this a particularly advanced capacity, yet AI systems lack it because the way we create them presently requires them to know the entire aim. If we construct systems that are aware that they do not know what the goal is, they will begin to display certain behaviours, such as requesting permission before removing all of the oxygen from the atmosphere.

Control over the AI system stems from the machine's ambiguity about what the ultimate goal is in all of these ways. And it's when you construct machines that are convinced they have the purpose that you have this kind of psychopathic behaviour.

I believe we see the same phenomenon in humans. What happens when general-purpose AI is introduced into the actual world? How do things evolve? Can we change?

This is an old topic. Surprisingly, Aristotle had a section in which he says, "Look, if we had fully automated weaving machines and plectrums that could pluck the lyre and produce music without any human intervention, we wouldn't need any workers."

Unemployment due to technological advancements

People understand that concept, which I believe Keynes coined as technological unemployment in 1930. They believe that if the machine does the work, they will be out of employment. Consider the warehouses that corporations are already using for e-commerce; they are 50% automated.

The way it works is that in an old warehouse, where tonnes of stuff is piled up all over the place and humans rummage around and then bring it back and send it off, there's a robot that goes and gets the shelving unit that contains the thing you need, but the human still has to pick the object out of the bin or off the shelf, because that's still too difficult.

But, at the same time, would you build a robot that is accurate enough to pick almost any product from a vast array of objects available for purchase That would result in the loss of 3 or 4 million jobs all at once? E.M. Forster penned an intriguing novella called, Machine dependent civilisation where everyone is completely reliant on machines.

The story is basically about how, once you pass over control of your civilization to computers, you lose the incentive to comprehend it or teach the next generation how to grasp it.

"WALL-E" can be seen as a modern version, where everyone is weakened and infantilized by the machine, which has not before been conceivable. Books contain much of our culture, but they cannot operate it for us. As a result, we must always teach the following generation. It amounts to roughly a trillion person years of teaching and learning and an unbroken chain stretching back tens of thousands of generations.

What will happen if that chain snaps? That is something I believe we must grasp as AI advances.

AI for all purposes

You won't be able to pinpoint the exact day of the arrival of general purpose AI because it isn't a single day. It's also not a case of all or nothing. The influence will grow in the coming years.

As a result, each advancement in AI broadens the variety of tasks dramatically. In that sense, I believe most experts believe we will have general purpose AI before the end of the century.

The median date is around 2045. I lean towards the conservative side of things. I believe the problem is more difficult than we realise.

I like what John McAfee, one of the AI founders, stated when asked this question: "Somewhere between five and 500 years." And, I believe, multiple Einsteins will be required to make it happen.