The Shadow Side of AI Part 3: Let's play sci-fi

What is the reality behind the AI headlines? Are we really headed for a dystopian nightmare or are we simply being taken for a ride?

By crossing into a space whose curvature is no longer that of the real, nor that of truth, the era of simulation is inaugurated by a liquidation of all referentials…It is no longer a question of imitation, nor duplication, nor even parody. It is a question of substituting the signs of the real for the real.

Jean Baudrillard - Simulacra and Simulation

Let’s play sci-fi - Image by Stable Diffusion

I spent the last couple of weeks down a rabbit hole that made me feel like I had entered Jean Baudrillard’s mind. If you have read any Continental Philosophers from the last 100 years, you often have to grapple with their words for some time to gain any sort of clarity into their meaning. It’s a bit like Brazilian jiu-jitsu for your mind. It all started with an article from Time magazine listing the “Top 100 people in AI”. I started going through this list of Top People, with a plan to listen to and read the views they held on the topic of AI risk. Frankly speaking, what started to unravel knocked me off balance. The more opinions I took in, the deeper I was led into a strange liminal space. As I looked around, everything seemed familiar, but the words I was hearing had started to take on a dream-like quality. Warning signals started to fire off in my head.

If I take what many of the ‘Top people in AI’ are saying at face value, I must confess that the tech world has become unhinged from its scientific roots and entered the realm of science fiction fantasy. As I listened to some of these fantasies unfold, the stench of bullshit was so great, that it made me feel quite queasy. I knew that chancers like Elon Muck, were peddling sci-fi futures about robots taking over the world and the need to fly off to Mars to save human civilization. That’s Elon Muck’s altruism for you. That’s his brand. Fabricate a dangerous enemy and position yourself as the perfect hero to overcome it. School boy fantasy stuff, the hero’s quest. But as the richest man in America he can play out his fantasies for the rest of his life. What more can we expect from a  man whose self-confessed philosophical tenets come from the Hitchhiker’s Guide to the Galaxy or Isaac Asimov’s Foundation Series. I mean really. Really.

It is one thing for the likes of Elon Muck to go off on his sci-fi trips, but what I didn’t quite realize was the extent to which these ideas had taken root outside his sphere of influence. Or perhaps I naively didn’t consider the extent of his influence. Of course, I knew that when he bought Twitter, one of the first things he tasked its engineers to do, was to tweak the algorithm so his Tweets were amplified by a factor of 1,000. But having never Twittered, I perhaps didn’t understand the raw power that this has put into his hands. If billionaires like Elon Muck can afford to buy platforms like Twitter as a personal loudhailer with scant concern even to make it profitable, then what chance is there for quiet voices of reason to be heard over the deranged outbursts of people like him? Trapped as we are in the calculus of the hyperreal media circus, attention grabbing flourishes will always have more travel than insightful facts.

But what knocked me off balance, more than anything, is the shift in power dynamics that are on clear display. It seems like the normal institutional boundaries are unraveling before our very eyes. A liquidation of all referentials as Jean Baudrillard put it. The fact that these tech bros seem to have the political leaders of the western world on remote control, ready to jump when asked to jump, is quite astonishing. Politicians are not voted in for their intellectual prowess, but watching the likes of British prime minister Richboy Sunak, demur so obsequiously to Mr. Muck shows the extent to which Big Tech has a stranglehold on the corridors of power.

As far as those other ‘Top people in AI’ who also choose to play along with Elon’s sci-fi visions, well perhaps they are just being pragmatic. Aside from the obvious commercial interests shaping their words, they may be finding some comfort by pointing us towards some fantasy realm and directing our attention away from the real issues we are facing. They may be keen to obfuscate the risks that the current set of Generative AI tools have brought with them, as they relentlessly rush to market, rolling out new tools to the masses on a daily basis. As I stepped back from some of these ‘Top people in AI’, I had to check myself quite a few times. To find my bearings, I looked around to see what others were thinking on this matter.

To my dismay I found ‘leading philosophers’, ‘historians’ and other ‘public intellectuals’ similarly willing to leave their critical minds behind, only too happy to add to this surreal moment and amplify the torrent of sci-fi propaganda even further. I am reminded of the experiments on social conformity by Soloman Asch who stated:

"[The fact] that intelligent, well-meaning young people are willing to call white black is a matter of concern."

Fortunately, you can find a few voices of sanity out there, even amongst the Top 100 in AI, if you look hard enough. Often marginalized or underfunded, these voices at least gave some credence to what I was experiencing:

The [idea of the] advent of artificial general intelligence, the surpassing of humans by machines, is really an article of faith, there’s really no evidentiary basis for that, it is just asserted by people who have enough social capital and enough power to be able to make assertions like that, and not be held accountable for them.
Professor Lucy Suchman – AI Now Salons

“When people hear our thing will become superintelligent,
smarter than people, people won’t be needed any more,
people might be served if we are lucky. If the machines don’t kill us.
But we won’t be needed anymore because the machines will be better at everything. It’s a grotesque lie.“
Jarod Lanier, Lead scientist, Microsoft Corporation

I have been a fan of science fiction since my childhood. The key word there is fiction. I first studied artificial intelligence over 30 years ago. The key word there is artificial. “Artificial Intelligence” simulates activity or processes that we would normally associate with intelligent living beings. To do this, humans program machines that are built by humans, to model and simulate the things we do. These simulations can only ever do what they are programmed by us to do. The idea that these programs could suddenly burst into a life of their own is a ridiculous category error.

Despite marketing claims that the current crop of AI applications have some sort of “reasoning” or “intelligence” of their own, ChatGPT and other Generative AI work in reality just like most other data processing systems out there. You pass it a prompt as an input, and it provides appropriate textual output. Any semblance of intelligence you impute onto this textual output is simply down to the labor of countless unnamed humans involved in the data processing supply chain. The idea that such systems could evolve into a form of super intelligence is no more than wishful thinking.

It's human labor, all the way down

Firstly there is the great data heist, where the work and efforts of millions of human creatives is freely scraped from the web, without permission, attribution, or compensation. This source data is then cleaned up with more human labor to shape it into a form that can be processed by the algorithms. This cleaned up data is then parsed using the brute force of huge computing power and resources to gradually build up a complex data structure that mathematically models patterns and other attributes in the input data. These models and algorithms were conceived of and programmed by human scientists and programmers. As end users we don’t get access to the raw unaligned model produced during this initial build. The AI companies make huge efforts to provide a censored ‘politically correct’ view of these raw data models using a final stage of tuning called reinforcement learning with human feedback (RLHF).

The RLHF process adjusts a model into a form that is palatable to contemporary mores and ensures it aligns with the ideological views of the AI companies producing it. Completing this stage is an often traumatic, manual process that ensures the most egregious, dangerous, or biased content scraped from the darkest corners of the internet is filtered out of the model. This task can’t be automated as it requires the subjective judgement of real humans, so more intensive human labor is recruited to do the job. Such dirty work is typically outsourced to developing countries where lax labor laws can be exploited for profit. Polished, tuned and fully tested by humans the model can finally be released to a hungry human public.

As I reveal the human efforts behind the veil of Generative AI, my mind tends towards the futility of my words. Many observers have revealed the environmental damage caused by the fast fashion industry as they create their wares in sweatshops far away from the fancy storefronts. These revelations don’t really dent the demand for cheap clothes. It just becomes another public relations problem for corporates to solve. Likewise, framing the efforts of humans as the real cause for the simulated intelligence provided by these Generative AI tools may be technically accurate. But it doesn’t really stop people wanting to believe in their sci-fi fantasies.

But just as all semblance of intelligence comes ultimately from human sources, likewise all risk comes from humans too. As former hacker and security expert Georg Hotz recently told Lex Fridman on his podcast:

“These large language models are good at fooling people because they were trained on a whole bunch of human data and were told to mimic it. I am so not worried about the machine, independently doing harm. This is sci-fi B movie garbage. The problem is bad humans using the unaligned AI to do whatever they want.”

If you have a few basic computing resources, you can plug into these Generative AI algorithms and create plausible text, images and even videos at an industrial scale. This is a gift to any bad actors out there. The power to rile up, catfish, and manipulate people at a more granular level than ever before is now on the table. As hinted at by the current Chair of the Federal Trade Commission, Lena Khan, Generative AI systems provide huge potential for bad actors to “turbo charge fraud”. The algorithms supplying us with news and information have already polarized us like never before. With Silicon Valley giving access to these Generative AI apps freely, who even needs external enemies like China? Elections are looming on the horizon. Let’s play sci-fi.