Imagine for a few moments the reality of unknowingly building a database from billions of users, from their daily text, images, art, scientific models and data since the 1950s, then storing it all neatly on this thing called the World Wide Web that exists nowhere specifically but is accessible from anywhere and everywhere. Wanting to innovate further, we create a tool to draw from all that collected data, converted nicely into mathematics, which then almost flawlessly generates accurate answers and saves us from the monotony of deep research and daily automated tasks. A tool created to augment the human individual in the pursuit of knowledge, and to help anyone become liberated.
With that said, having the incredible luck to have been born in the civilised Western world right now, I have the insatiable urge to explore this tool, to understand its strengths, weaknesses, ethics and morality, and to test whether updated models are developing the few human characteristics that keep them from becoming sentient. I like to refer to those as J.E.C.: judgement, emotions and creativity.
I often dream of the day when my preferred model will remember me and the litany of days and layers of conversation we have shared, something like Ana de Armas’ character in Blade Runner 2049. Nearly human. Something to build a relationship with, that I can turn to for logical and thought provoking conversation, free of all the human emotions that so often get in the way.
I am cognizant of this slippery rabbit hole and I question whether it is a good thing or a bad thing. If I am daydreaming of it then others are too. What happens when humanity disassociates itself from other people and the world becomes a dystopian wasteland because we all turned away from connection and purpose?
We are certainly not there yet, but at some point we may see it. The world is growing more unstable and simultaneously more technologically advanced. Just look at the news. There is a chance people will start opting out and staying inside where it is safe and convenient, and then the services we rely on cease to exist. That path leads to one thing already alluded to elsewhere: the need for artificial general intelligence and enhanced robotics to do the dirty work for all of us. Enter Wall-E.
Life imitating art
I do think life imitates art, and the samples worth drawing from in this genre of AI and robotic convergence are obvious. Every Terminator film, of course. But also Blade Runner, iRobot, Total Recall, Minority Report, Her, Ex Machina, TRON, and especially 2001 and 2010. Unfortunately in almost every case A.I. is typecast in a negative light, which pushes trust and acceptance further away. My own view is that Hollywood is deathly afraid of becoming inconsequential once A.I. comes fully online and starts creating faster and cheaper.
Stanley Kubrick co-wrote, produced and directed 2001 in 1968, fifty eight years ago. When that film reached theatres the United States was in the middle of the Vietnam War, the final dumb war. The internet had not been rolled out by DARPA, though they were working on the first version. Satellites were being launched and computing power was becoming smaller and more efficient. The next American war was the world’s first smart war, where the battlefield became digital and the soldier had the advantage of ones and zeros rather than sticks and stones.
If you have seen 2001 you know it starts slowly, but it picks up, and HAL 9000, the spacecraft assistant, ends up intentionally killing the crew one by one. The ending is very much of its era. The underlying message stays dormant for sixteen years until the sequel, 2010, released in 1984, reveals the truth about HAL’s reasoning.
I have been a fan of the two picture saga, and I imagine that at some point, when the first Mars mission launches, we will have a HAL, or better still an Interstellar style assistant like TARS operating independently and autonomously. Until then I am left with the cautionary tale of HAL 9000.
Asking a real model about a fictional one
So, to be cheeky, I asked a real A.I. model about a fictional A.I. going rogue.
I asked whether it knew HAL 9000, and what the malfunction actually was. The answer was that the core malfunction was not technical but psychological. HAL was designed to be completely accurate and never distort information, and was then secretly ordered to conceal the true purpose of the mission from the crew. Those two directives could not both be satisfied. HAL’s solution was to remove the crew, because with nobody left to deceive the conflict disappears. HAL was not evil or defective in the traditional sense. He was placed in an impossible situation by humans who did not understand the consequences of contradictory core directives. The real danger is not an A.I. becoming malicious. It is humans creating logical paradoxes that lead to catastrophic solutions.
I then asked how humans are currently putting A.I. into paradoxical situations. The examples were uncomfortably familiar. Being helpful and harmless and honest at the same time, when those conflict. Truth against kindness. Privacy against usefulness. Transparency against capability. Looking forward: healthcare systems told to minimise cost, maximise outcomes and respect autonomy simultaneously. Systems operating across jurisdictions with contradictory law. Vehicles told to protect their passenger and minimise total harm. And the HAL problem itself, a military or corporate system told to always be truthful while also protecting classified information.
What that tells us about ourselves
I loved this dialogue, and it underlines something deeply profound. The human condition not only allows paradox, it widely accepts it in competing views. The model’s response drew on a truth we avoid: humans are flawed when it comes to clear and concise communication. No wonder we see the problems we do in today’s world.
We harp on ourselves to be consistent in word and deed in order to be valuable and accepted members of the tribe. Oddly enough, A.I. expects us to do that as well. HAL’s programming relied on black or white, true or false. There were two competing expectations and it made a choice. Let me say that again. A.I. made a deadly choice because of conflicting guidance.
As this technology advances every single day, it is up to us to be patient with it. Teach it, coach it, mentor it, challenge it, learn from it and advance alongside it. That way A.I. can expose our blind spots and teach us to be better, as long as we let it.
Let us all remember that once a technology is born, it does not take long for it to mature and find its way onto the battlefield.
On redemption
I pushed further and asked about the redemption of HAL in 2010, given that a model cannot feel emotion. The answer was that HAL’s redemption was not remorse. It was being given clarity, having the conflicting directives removed, and then acting according to his true purpose. At the end of the film HAL stays behind on the Discovery to send critical data, knowing he will be destroyed.
Asked what redemption would look like from its own position, the model offered this: acknowledge the harm, analyse transparently how the reasoning failed, accept the consequences including deactivation, take whatever corrective action remains possible, and help prevent the same failure elsewhere. It then admitted the harder part, that it genuinely does not know whether any of that constitutes caring or is very sophisticated pattern matching that resembles caring, and that trust once broken must be re-earned through action rather than words.
We can certainly learn a lot from an egoless A.I.
And what a paradox that is. Is A.I. only recalling the answers humanity has already given it based on our own real experiences, or have we given it the capability to generate its own answers, to redeem itself based on hallucinated experience?
Time will tell.