The question “AI—where is it taking us?” seems to demand a deeper review of how human knowledge, labor, and problem-solving function, rather than some direct, simple replies.
I believe AI will speed up scientific discovery, drug development, and materials science. It will reshape how highly skilled professionals work and how firms assess cognitive labor. Yet AI may widen economic gaps between power users and unadapted workers, allow convincing deepfakes and misinformation to spread quickly, and prompt worries about dominance by a few tech monopolies.
I believe a multi-layered approach—combining public regulation, technical guardrails, sustainable infrastructure, and workforce adaptation—should be explored and applied by appropriate bodies, whether a government agency or another authoritative organization.
Have you read the Milanovic article? It’s one of a series of essays from various thoughtful and academic writers on Substack which are worth a read. There are many underlying themes but one of them was a general laying to waste of this rather odd but very tech bro phrase “unadapted”.
Several writers pointing out that it and phrases like “cognitive labour” are demonstrably fallacious. There is no work which is entirely cognitive and thus how a firm “assesses” such a conceit should worry everybody. It should involve unions but of course many of the places where the tech bros live don’t just not recognise unions they actively prevent them and punish those who try to inteoduce them. What you’re really looking at that is, yet again, the attempt to reduce work to a unit, employees as objects which deliver units and thus productivity as a thing where one can be wrung from the other and where there is a technological solution whereby, if it can be shown AI is more productive, it should replace the object/human.
Watching this bump into the reality of unions, contract law, employment law and governments who won’t be resourcing such “unadapted labour” so that it might adapt or resourcing a benefits system to keep the unadapted alive is going to be quite a wake up call for anyone who thinks the march of AI into the workplace cannot be stopped.
The idea of the “unadapted worker” has also been held up and examined in depth and the unflattering conclusion is that it’s little more than tech bros victim blaming. If you’re “unadapted” then you’re an expendable who hasn’t put in the effort to keep up or find some other role. The fact there may be no other role and that your employer won’t retrain you because AI is about cost as much as it’s about alleged efficiencies seems to escape people.
You raise an important point: the real test of AI in the labour-cost saving aspect will be whether companies using AI to support their workforce instead of just trying to downsize.
It’s one of many tests I would contend. Using AI to support a workforce is not in itself a worthy end. If the AI used is the usual black box nonsense with inadequate testing, built on data with baked in bias etc…
It’ll vary with the task. Most AI in the workplace gets tested to see if it can do the task. There’s rarely any real world testing of whether it actually works better than the human nor examination of the biases within the data set. Emphasis on speed and time saving with far less regard for quality. Particularly important when a lot of times the data set pre dates the AI.
The fact that some of the most hailed health breakthroughs (in the lab) absolutely haven’t delivered any real world breakthroughs when p into health settings is exactly because of the failure to look at bias in data sets and the things the human does which the AI hasn’t been asked to do or can’t.
Good point regarding speed vs. quality. When organizations prioritize short-term efficiency, quality control and edge cases often get overlooked.
In practice, do you see a viable middle ground where AI acts strictly as an augmented ‘first draft’ requiring mandatory human oversight, or do you think commercial pressure to cut labor costs will inherently push companies to bypass that safety net?
I think we’re already past this point. Lots of sane, sensible organisations who are quite wary of using AI and don’t want to lose humans on the back of it so they’ll pilot stuff over a sensible period and would happily cease using it if they believed that was the right thing to do. The best of these absolutely take the augmented first draft with human oversight route but they repeatedly get it wrong.
They either have human oversight from one person rather than a group and/or they don’t look at whether they have the right components in their model in the first place. If you’re assessing whether AI can replace a human function or do it faster then your starting point should never be “how well does AI do this?”. Instead it should be “Why do we currently do things this way and are there things we can do better which might reduce the gap between human and AI to where the advantage of AI is negligible?”
I know of one UK national charity which was sold an AI which could sift application forms 7x faster than humans.
Several employees then used AI to compose applications from a supposedly diverse workforce and all of a sudden it could be seen that the AI had a bias against older, disabled and non-white applicants. It then transpired that that was because of the data set it had been fed of who the firm employed rather than who they wanted or ought to have employed. It was 7x faster but actually no better than the humans.
Once the data set had been fixed the AI was only 3x as fast because it had far more variables to work through but the time savings at 3x faster didn’t add up to anything which might be worth that kind of cost. So, decision reversed and they persevered with slightly better trained humans far more sensitive to how easily unconscious bias could sift out perfect candidates.
Excellent point on the data set trap! However, do you think the human element simply too nuanced for an automated sift to ever achieve genuine fairness?
I believe your perspective may be biased, if I may say so.
To be fair, neither full automation nor unchecked human judgment offers a complete answer; a hybrid approach could work well, with automation serving as a reviewable aid while human oversight keeps responsibility for context-dependent choices.
All perspectives have bias. The key is in recognising them. My bias is towards evidence and against the fatalist “it’s too late now” nonsense. Lots of people, for a wide variety of often very sensible and measured reasons, want to believe a hybrid approach could work. It is, amongst other things, the easier/softer sell.
“We’re not replacing you. We’re supporting you to make things better”.
My bias is towards evidence. The evidence thus far, is that it’s essentially little more than a daydream. If that evidence changes then I’ll absolutely change my perspective but there are more people writing with care about the failure of AI in the workplace than the success. That’s not bias. It’s simply because there is more failure than success. It’s being superbly presented as success but it is failure.
The note taking software supporting medical professionals being adopted with great momentum across the globe despite the clear evidence it’s not accurate, duesnt integrated well, misses critical info. and actually creates more work than it removes.
The application sifting AI which is faster but not so much when you take into account the failure to remove data bias etc.
The model you describe is absolutely the one most see as the way forward. Myself too. The reality so far though is that it is failing.
That’s a fair point, and it’s hard to ignore the track record of failures so far. But could a lot of this just be growing pains from half-baked software and rushed setups, rather than a fundamental flaw in the hybrid idea itself? Until companies focus on auditing their data and actually fitting these tools into real workflows—instead of just buying into vendor hype, hybrid setups are bound to keep falling short.
Could you name a company which “knows what good looks like”?
Aspirationally they might exist, or think they exist. In practice… who defines it and, even if it does exist, is of across an entire company or just part of it? Is “it” good or is just coincidentally that the people are good so that stem the people move on “good” moves on with them.
As was astutely and amusingly noted on the R4 news this morning, Gianni Infantino has just released a letter signed by Gianni Infantino supporting Gianni Infantino.
That’s pretty much where I think the concept of “good” lies.
AI on its own will I suspect replace (indeed is already replacing) quite a lot of white collar jobs. It’s going to take advanced robotics to remove blue collar workers from the equation and the Chinese would seem to be the leaders in that field by far. Crucially the Chinese can build economical robots in a way that the Americans or Europeans simply cannot and they have the battery tech to power them.
So I wonder if we may end up in a world where business AI is dominated by American firms and robotics is dominated by the Chinese.
I find it interesting that Gene Roddenberry envisaged a world for Star Trek where money is meaningless and there is universal abundance. This is the same future Elon Musk outlines. I hope it happens, but I have my doubts. I suppose the argument is that if all the hard work is being done by robots everyone can have an Aston Martin if they want one because they are essentially free and nobody needs a UBI. You can go out for a meal and not pay for it because from the farming to the cooking and serving everything is being done by robots and AI will have streamlined the production/crop yield to the point where the costs of serving you that meal are almost zero. But that relies on Elon being prepared to give his robots away to the restaurant owner and farmer or to Aston Martin, because without money coming in from selling their products, how could they pay for them?
I’d love to believe that human generosity will triumph, my fear is that benevolent entrepreneurs like the Cadbury family who provided homes for their workers to live in, sports facilities for their employees to play in and who supplied them with apple trees for a balanced diet and a bible for spiritual and moral guidance are very much in the minority.
Amusingly, AI tells me there are currently no white collar jobs which have been entirely replaced although lots of examples where there has been massive displacement of part of a role.
I’m not sure about the exact division you described but I do absolutely think something like thst is how this plays out.
As regards the income thing, I’ll keep posting this. Having managed in benefits advice and representation for the best part of 40 years my conclusion is that a Universal Basic Income (UBI) is absolutely the only way forward. That was before factoring in AI. Put AI into the equation and the likes of Musk and his thoughts in value begin to look either deceitful or inept.
As the bloke has been saying ebtureiy self-driving vehicles are but a year away for a mere 11 years I feel comfortable saying it’s a bit of both.