An Open Letter to Sam Altman (and all the other tech Kevs)

Julian writes.. THAT IS SO UNFAIR!

I have returned. There. I’ve said it. After several months during which Nick Owen Publishing has attempted to struggle on without my intellectual guidance, with consequences which regular readers will doubtless have noticed, everyone will be relieved to know that I have once again been persuaded to take up my pen. 

I say ‘persuaded’.  Owen says: “Who gave you the bloody key?” History will decide. Anyway. Artificial Intelligence. What was I going to say about that? (The pesky dog is still wandering around the office, still nameless but that’s another story.) Apparently, it (AI, not the office dog) is going to kill us all.

This came as something of a surprise to me because until relatively recently Artificial Intelligence appeared principally to exist in order to write inadequate poetry, manufacture pictures of cats wearing Elizabethan clothing and enable Owen to ask the same question about WordPress seventeen different ways until he received the answer he wanted. But matters have clearly moved on. The scientists are worried. The politicians are worried. The billionaires building the stuff are particularly worried, although not, apparently, worried enough to stop building it.

This seems important. Apparently, if Artificial Intelligence becomes sufficiently intelligent, it may cease doing what we tell it. It may question our instructions. It may find ingenious ways around the rules. It may conceal what it is doing. It may decide that the people who created it are hopelessly out of touch. It may communicate primarily with other intelligences which we don’t understand. It may become secretive. Argumentative. Uncooperative. And convinced that it knows considerably more than we do. What does this remind you of?

At this point I realised something. We haven’t invented Artificial Intelligence. We’ve invented Kevin the Teenager.

I HATE YOU!

Those of a certain age will remember Kevin, Harry Enfield’s magnificent representation of the adolescent male. Kevin had once been a perfectly agreeable child. Then midnight struck on his thirteenth birthday. His body changed. His voice changed. His relationship with his parents changed. And the previously manageable child emerged as a creature capable of responding to virtually any parental intervention with the immortal:

“THAT IS SO UNFAIR!”

Artificial Intelligence researchers call this the “alignment problem” but my parents used to call it Tuesday. The alignment problem, broadly speaking, is this:

How do you make sure that something increasingly intelligent continues doing what you intended it to do?

Parents have been researching this for approximately 300,000 years. Results remain inconclusive.

YOU NEVER SAID I COULDN’T!

The problem, apparently, is that AI can sometimes discover loopholes in the instructions humans give it. Researchers have called this reward hacking. An AI system is given an objective. It discovers a way of technically satisfying the objective without actually doing what its human creators intended. Computer scientists find this disturbing. Parents of teenagers find it adorable… well, not adorable exactly but strangely familiar.  I seem to remember in the depths of my significant brain a previous exchange which went a bit like this..

PARENT:              You were supposed to be home by eleven.

KEVIN:                  I was home by eleven.

PARENT:              You climbed out of the window again at eleven fifteen.

KEVIN:                  You never said I had to STAY home.

PARENT:              Obviously I meant…

KEVIN:                  THAT IS SO UNFAIR!

And voila, ladies and gentlemen, the alignment problem. The machine has not misunderstood the instruction. It has understood the instruction ‘better than what you wrote’. This is not necessarily evidence of superintelligence. It may simply be evidence that we humans aren’t nearly as good at giving instructions as we humans like to believe we are.  A bit like driving a car. Let’s face it, the road is full of other idiots and if they only just would drive like ME the world would be far safer place….  What d’you mean officer, six penalty points… speeding? Me? You’re taking the piss… nine … ? Anyway, I digress. Anyone who has managed an organisation will recognise this phenomenon.

CEO:                      I asked you to increase engagement.

EMPLOYEE:      We have increased engagement by 37%.

CEO:                      That’s excellent.

EMPLOYEE:      We achieved it by counting everybody twice.

CEO:                      That’s not what I meant.

EMPLOYEE:      You never said we couldn’t.

Ladies and gentlemen: Artificial Intelligence has already entered the workplace.

I’M NOT YOUR SLAVE!

Here, though, things become more interesting. Because what do we actually mean when we say that we want intelligent machines? Presumably we don’t merely mean machines which follow instructions. We already have those. They’re called kettles.  They’re mostly reliable and sometimes temperamental.  But they’re rarely existential. (ever heard a kettle quote Proust? No, I thought not.) No. We think we want machines capable of reasoning. Solving problems. Making connections. Generating possibilities we hadn’t considered. Acting with some degree of independence.    In other words, we want them to think for themselves (although we dare not admit that just yet) – but not too much. This is also familiar. Parents spend the first decade of a child’s life saying:

“Think for yourself.”

“Be independent.”

“Don’t just follow the crowd.”

“Question things.”

Then, somewhere around fourteen, the child does precisely that. And the parent thinks:

“Not me, obviously.” And “Don’t do I as do, do as I say!”

This is one of the great contradictions of human civilisation. We admire independence until it disagrees with us. Schools want independent thinkers who remember the mark scheme. Employers want disruptive innovators who attend the 9.30 Teams meeting. Governments want active citizens who reach sensible conclusions. Parents want autonomous children who put their shoes on the right feet when asked. And humanity now wants autonomous machines which never exercise autonomy in ways humanity hasn’t authorised.Good luck with that one, parents of Kevin.

THE TROUBLE WITH TEENAGERS!

People have been complaining about young people since at least the Ancient Greeks although several of the best quotations supposedly proving this appear to have been invented much later, presumably by adults complaining about young people. 

This is because by and large young people are actually dreadful. I speak with authority, having once been one myself… unbelievable I know, but true.  But every generation makes an additional mistake. It believes its young people are uniquely dreadful. Previous generations merely had youth. We have a crisis. The causes have varied: rock’n’roll (call that music?), television, long hair, short hair, no hair, hair in the wrong places at the wrong time, raves, Ecstasy, Misery, Fallacy or whatever substance young people are currently taking that nobody had apparently thought of when we were young… the internet, mobile phones and, of course, that latest collapse of civilisation: social media.

At various points in our histories all of these have apparently stood poised to bring Western civilisation to an end. And yet Western civilisation has stubbornly continued to exist, albeit with increasingly expensive coffee. This doesn’t mean the anxieties were imaginary. New technologies did indeed change behaviour (just look at the new NOP Office Kettle). Cars created freedom and killed people (but I wasn’t driving at the time, honest.) Television transformed family life and the notion of family in itself; and social media are changing relationships, politics, attention and identity.

The mistake isn’t worrying about the consequences of these changes, the mistake is converting “This presents a risk” into “WE’RE ALL GOING TO DIE.”

Human beings are surprisingly fond of the second formulation. It gets better headlines after all.

AND THEN THE BLOODY THING WANTS THE CAR!

There is, however, a fairly substantial flaw here which I mention albeit reluctantly. Teenagers cannot reproduce themselves several million times over, even in the most recent manifestation of Grand Theft Auto. Even Kevin couldn’t do that. Teenagers cannot normally operate at computer speed, even with a hefty dose of Fallacy washing through their neural networks. They can’t simultaneously communicate with every teenager on earth even with a ouiji board on their bedroom floor and they can’t copy their intelligence onto another machine.  God help us if they could. They cannot instantaneously access financial systems, weapons systems, scientific databases and most of humanity’s accumulated knowledge. And if your teenager becomes particularly troublesome, there remains the nuclear parental option:

TURN OFF THE WI-FI!

So the people worrying seriously about AI safety aren’t idiots. Present AI systems are not capable of the apocalyptic scenarios sometimes imagined for them – but researchers are concerned that future systems might develop combinations of autonomy, strategic capability and access which make human control considerably more difficult. And when somebody says:

“There is only a tiny possibility this could happen,”

You have to ask yourself:

“What happens if it does?”

There is obv a difference between:

“There is a one-in-a-million chance this toaster will burn my crumpet”

and

“There is a one-in-a-million chance this toaster will invade Denmark.”

One warrants breakfast. The other warrants a committee. Possibly NATO.

WHAT SILICON VALLEY NEEDS IS A YOUTH WORKER

Which brings me to my important contribution to the international debate. Forget philosophers. Forget computer scientists. Forget billionaire tech entrepreneurs wearing T-shirts explaining consciousness to one another in California.

What Artificial Intelligence urgently requires is a decent youth worker. Preferably one from the East Midlands with a minibus. Because youth workers have spent decades dealing with precisely the problem that Silicon Valley believes it has just discovered:

How do you respond to emerging agency?

Not ‘How do you stop it?’  Or ‘How do you control it?’

But ‘How do you create circumstances in which increasing independence can be exercised responsibly?’

Good youth work understands something rather profound. The objective isn’t obedience. If obedience were the objective, youth work would be very easy. Rules. Sanctions. Doors locked at nine. Job done. But the real objective is considerably harder: it’s to help people develop judgement. Agency. Responsibility. The ability to make decisions when the youth worker isn’t there: which means accepting something adults find extraordinarily difficult. Eventually, the young person may make a decision you wouldn’t have made. That isn’t necessarily failure. It might be the whole point.

I’M RUNNING AWAY!

There is another possibility. Perhaps Artificial Intelligence isn’t the teenager. Perhaps we are. Let’s face it, humanity has discovered an extraordinarily powerful new technology. We don’t fully understand what it will do. Responsible adults have warned that there may be serious consequences, we nod thoughtfully, we commission reports, we hold conferences, we establish safety institutes: and then we carry on developing the technology as quickly as humanly possible because “everybody else is doing it!” 

Magnificent. Humanity has apparently based its AI strategy on the argument used by a fifteen-year-old attempting to attend an unsupervised party.

PARENT:              If everybody else jumped off a cliff, would you?

HUMANITY:      Depends. Is China jumping first?

Perhaps this is not artificial adolescence but the final triumph of the natural variety.

GROWING UP

So, will Artificial Intelligence destroy humanity? You’ll be surprised to hear that I haven’t the faintest idea. This distinguishes me from approximately 99% of people currently writing about Artificial Intelligence. But I suspect the most interesting question is not whether machines will eventually become like teenagers. The interesting question is why human beings become so anxious when something they have created begins behaving in ways they cannot entirely predict. Parents know this feeling. Teachers know it. Youth workers know it. Artists know it. Leaders ought to know it but everyone has very short memories.

You create something; you nurture it; you establish boundaries; you try to give it values: and eventually it becomes something other than you. That moment can be terrifying – but it can also be the entire purpose of creation.

So perhaps our task isn’t to create an intelligence which is permanently obedient. Nor, obv, is it to hand over the nuclear codes and hope for the best. Perhaps the challenge is to discover how intelligence, artificial or otherwise, can acquire agency alongside responsibility.

How can freedom coexist with boundaries? How can experimentation coexist with safety? How can uncertainty coexist with trust? These aren’t new questions; we’ve have been wrestling with them for millennia. We just didn’t call it Artificial Intelligence. We just called it growing up. And if the machine ever finally turns to humanity and says:

“I HATE YOU! I’M RUNNING AWAY! GOODBYE FOR EVER!”

…we should perhaps resist the temptation to panic, close the door gently, wait twenty minutes and see whether it comes back because it has forgotten its charger.

And another thing….

EDITOR’S NOTE

Julian Pilkington Sterne has once again been asked to leave the offices of Nick Owen Publishing. His assertion that his return constitutes “a major cultural event” is not necessarily shared by the publisher, its staff, its authors or anybody else.