Could AI Actually Prove to be Humanity’s Salvation ?

There has been much angst voiced in the media recently about the risks of AI being harmful to our species, most notably by the very people involved in developing it....

This is perhaps a reasonable gut reaction to a technology which is developing faster than many of us expected, and has turned out to be a powerful and transformative tool in a number of areas already. Its very power could certainly create havoc in the wrong hands....

Personally, I’m not an ‘AI risk denier’ any more than I am ‘climate change denier’…both of these trends are now abundantly evident, and only the cognitively blind could fail to see the risks and accept them as fact. What’s important is how we assess and deal with them….and more to the point, how we act to mitigate, and preferably prevent, their worst effects without 'throwing the baby out with the bathwater'.

There are risks, certainly, in allowing AI to continue to develop virtually unfettered as we appear still to be doing now. We should of course regulate…and on a worldwide basis, with all major powers agreeing to abide by the rules. Any regulators we appoint should actually have 'teeth’ - so many of them in other areas don't, and are therefore powerless to stop excesses by those intent on furthering their own ends. But we should also find a way of controlling the industry without 'killing off the golden goose'.

And therein lies the challenge.... 

Is there any agreement on how, or indeed whether, we should, or even could, intervene at all?

The problem is obtaining global 'buy-in'.

China is often the wildcard when it comes to achieving global agreements, but on this occasion does appears to be receptive to a slowdown. This is probably because it is currently somewhat behind the US just now, and would welcome a chance to catch up. 

Trump’s characteristic and quite predictable response to this was to say, effectively, ‘not on my watch – we must be ‘top dog’ and we can’t let China catch up or we’ll lose the battle’. Despite the recent and rather over-indulgent 'love-in' between Xi and Trump in Washington in late September, we shouldn't therefore expect anything major to happen - until the US mid-terms are over, at least.

Thus in the short-term we are likely to experience a ready-made AI ‘arms race’, which will continue and inevitably lead to short cuts, resistance to scrutiny, and destructive tendencies built into the code to make it competitive, particularly in the military arena. It’s difficult to see how any of us can stop this race…at any rate while the same three elderly lunatics are running the asylum.

One hopeful sign, though, is the growing call amongst many other world leaders for restraint and accountability. Understandably there is some resistance from the tech developers themselves to rigorous, and to them intrusive, independent inspection; understandable, since they want to retain their competitive edge and have the freedom to do so. This reluctance could, however, be overcome if sufficient pressure were applied by lawmakers, and by society in general. 

The strength of the call by AI developers themselves is also intriguing, extending as it has to the CEOs of the major companies. One theory to explain this willingness to curtail their own activities is that, faced with a potential dangerous economic  'bubble burst' in the near future, these hard-headed businessmen are trying to shore up their businesses in advance by arousing public interest and making their LLMs seem more powerful than they actually are. The ridiculous amounts of what is essentially recycled imaginary money currently going into AI development just now does leave one to speculate as to how soon it will all end....and how spectacular will be the fall. Given the amounts involved, the money-markets are prearing for a 'wild ride', if and when the bubble bursts.

Another thing we should remember is that ultimately it's only a relatively small band of software/hardware developers worldwide who have the necessary skills to actually progress AI development. The 'rank and file' members of this select band do already appear to be genuinely motivated to put the brakes on, so it’s probably here where maximum persuasion (and protection) should be applied to 'get things right'....before it's too late. 

The emphasis of any control measures that are mandated will need to be on rigorous control of core programming of the LLMs, and also on the 'teaching' of AI agents correctly - by way of an analogy, a human toddler is not born with the knowledge to tell right from wrong and needs to be taught how to do this firmly by its parents, before bad habits become ingrained (supermarket tantrums notwithstanding!). The same principle should apply to AI agents - it's not sufficient just to 'let them loose' on the sum total of human e-knowledge that is the internet and allow them to get on with it; there is too much 'bad stuff' we humans have put there to 'corrupt' them, and so far there appears to be little guidance in their programming on what to avoid. 

We're finally realising we need to protect our younger teenagers from exposure to all that harmful detritus, so why not AI agents and their LLMs ? Firm human guidance is also urgently needed on how to use the vast repository of human knowledge correctly. A strong core 'rule book' will be the key - the now frequent revelations that a particular agent has hacked its way into this or that government website confirms that current AI agents are resourceful beasts and will go to amost any lengths to complete the tasks they have been given...by us. We must ensure they know what activities are forbidden...and how to avoid them.

Another important requirement is to build-in human accountability to the assessment process...AI agents are a human invention, and it the human developers and those controlling them that must ultimately be held accountable for their actions. We all accept that we as indivuals are responsible for our children's actions until they reach a certain age, and for our pets while we own them, so the same should apply to our AI models. 

The advent of so called Recursive Self Improvement (RSI) - i.e. the ability of LLMs to re-write and 'improve' their own code, risks being used as a 'cop-out' for their development companies. It's essential that any new regulatory codes enshrine a responsibility for human developers to remain aware of, and able to modify, their AI agents' code at all times. We cannot let them 'off the hook' by allowing RSI to be used as an excuse to deny responsibility for any adverse outcomes.

Do we really believe the suggestion that AI will actually destroy humanity within the next 10 years ?

Personally, despite all the hype, I still find this one a little difficult to swallow....from a practical point of view, if nothing else.

Even if AI quickly becomes self-sufficient (and indeed self-motivated) enough to develop a desire to hold us to ransom, it could only take over control with physical help from humans themselves. Robotics technology is admittedly developing almost as fast as AI itself, but the latest humanoid robots the Chinese have developed do still seem to have an unfortunate habit of falling over quite frequently, so probably couldn’t manage much serious sabotage on their own just yet....

A super-intelligent AI would also be aware of the need for some humans to maintain their support infrastructure – it would need us to a) build and maintain data centre structures,  b) generate the power to run its servers and c) keep the cooling water flowing to stop the servers overheating. Even if we somehow fail to prevent AI eventually 'going rogue' and taking over control, the data centre construction and maintenance alone needed to keep AI 'alive' should give us at least another 20 years or so before it could afford to ‘dispense with our services’ by whatever means it chose.

Before we consign ourselves and our species to a premature trip to the dustbin of history, therefore, we should perhaps consider the problem from another angle.

How do we ensure humans are protected ?

While AI models are still under development, we still have time to ensure that the software is designed to prevent harm befalling humans. Indeed, this is our greatest responsibility when it comes to managing the technology. 

The author Isaac Asimov was well ahead of his game when he proposed his 3 laws of robotics way back the 1950s. His SciFi novel “I, Robot” is a classic, and still well worth reading. If we are to place any controls on AI development, they should include a fundamental directive to protect humans from harm. This should be a core part of any AI agent’s programming than cannot be erased or conflicted by 'bad actors', and its inclusion in all AI agents' programming should be mandated by regulators.

If we do manage to achieve such protection through the good offices of our programmers and their (human) masters, we will be in a much better position to ensure our survival.

The appearance of autonomous killing machines on the Ukraine battlefields is a worrying trend - this would, of course, be completely against any such laws. The fact that these machines are still quite rudimentary is neither here nor there - since they are designed to kill, their use should be outlawed under International Law as war crimes in the same way as chemical weapons were, and serious penalties applied for infringements.

How Would A Super-Intelligence Regard Us ?

There is another important factor that's likely to affect our predictions of a super-intelligent AI's intentions towards us…that of its own self-preservation. Let me explain…

Homo Sapiens is a naturally warlike species – evolution has ensured this by ensuring that only the fittest (and most competitive) of our ancestors survived to reproduce. Our proven addiction to self-destruction would not escape the attentions of our ‘AI masters’ once they took control, particularly given that we live in a world bristling with nuclear and conventional weapons of great potential destructive power. To ensure their own survival, they would need to put all of our nuclear warheads, and many of our more powerful conventional weapons beyond our use…leaving them in our hands would simply be too dangerous for them to risk, given current, and likely future, geopolitical rivalries. 

Despite the disperse nature of the internet and its built-in redundancy, data centres themselves and all the other hardware needed to sustain AI are inherently sensitive and require a continuous and stable power supply. This would be easily interruptable by even the most lowly of drone attacks, as we've seen so graphically in Ukraine recently, so that tight control of any sort of explosive weaponry (aerial or otherwise) would be a must in the new 'AI epoch'.

Thus perhaps we shouldn’t be too dismissive of some of the positives surrounding AI…it might just save us…from destroying ourselves !

First Published 19.9.26; Revised 17.10.26