Hard thinking needed on how to control the elite behind AI: Andy Brown
Intelligence is a hard thing to define. Particularly when it is artificial. We are witnessing the emergence of machines that can do amazing things like translate at ease between different languages or replace the work of a highly-trained surgeon doing delicate operations.
We are also plagued by supposedly intelligent devices such as sat navs that still struggle to distinguish properly between a wide road and a narrow country track.
Or websites that fail to answer any complex question despite the company removing all phone numbers and email addresses from its contact us pages.
Artificial intelligence sounds very convincing but the machine gives out errors as confidently as correct information and lacks self-awareness. All living creatures have this. They know where their own existence ends and what is helpful to them or harmful.


We simply don’t know what it would take to transform an electrical device into a self-aware creature with flexibility to decide on its own motivations. We do know that this is something that is vanishingly unlikely to happen as a result of crunching huge volumes of data and detecting patterns that indicate what is most likely to be the most productive way to respond to a request.
There are huge advantages to be had from utilising the genuine abilities of machine learning. Most of them are best achieved in restricted and controlled areas of activity.
I would rather receive the opinion of a highly trained AI legal programme which can trawl every court case ever judged than that of a highly-paid barrister who may have very human gaps in their knowledge. I would also cautiously tend to trust AI to do accounts or suggest a likely medical diagnosis.
The important word here is ‘tend’. The oldest truism in computing is that if you put garbage in you will get garbage out. If a programme gives me a legal opinion I might want to be sure it had not relied on US law. If it wrote me an essay on the midterm US elections I might want to know something about the prejudices of the news outlets it was trained on.


As is so often the case with new technology the real challenge is not to find examples of how it can be helpful. It is to get used to knowing when it can’t be trusted.
There is no actual intelligence or understanding of the answers we are being given. There is only a bulk crunching of data that may contain some appalling prejudices or outright errors.
It is a hard enough challenge to sift through every offer of love and every offer to make you rich that comes routinely over the internet. It is going to become much harder to do that when those offers start coming from very friendly human sounding voices. Especially if that voice has previously supplied us with helpful information or has provided psychological support and friendship on demand.
There are some very unpleasant governments and ultra rich individuals out there who are putting a lot of thought and effort into how they can use the latest technology to manipulate us and undermine our way of life.
It is no longer necessary for Putin to bomb Britain to cause us huge damage. All he needs to do is employ programmers who force our supermarket freezers to turn their temperatures up and ruin food supplies as was done recently in France. Or hack our electricity grid and leave us in darkness. Or exploit the multiple weaknesses in our NHS computer systems, our school exam systems, our train services, or our weapons systems.
The biggest risk we face is not that a computer becomes sentient and decides humanity is a primitive nuisance. The biggest risk is that one set of humanity instructs it to do serious harm to another set and what it is asked to do gets out of hand.
The Russians recently had a case of the plague escape from a research lab. Despite all the experience with Covid we remain very badly prepared for an even worse pandemic.
There have been repeated cases of AI being given instructions which it then proceeds to carry out with ruthless efficiency that involve it hacking systems as it casually breaks the law and lies about what it is doing.
The key risk is not ultra intelligent evil machines. It is ultra stupid or evil people instructing them. AI puts dangerous amounts of power in the hands of bad actors.
The outcome of the war in Ukraine currently depends more on what decision Elon Musk makes about access to Starlink than what decisions the British Defence Minister makes.
It is not wise to give any individual such phenomenal power that they can decide the future of an entire country. The oligarchs and billionaires already control most of our energy supply, almost all of our conventional and online information sources, our satellite computer systems and the teams training almost every AI model.
We might be wise to put longer and harder thought into how we control the elite that is providing the inputs into AI rather than obsess about its independent development of intelligence. With the latest computers evil in gives you evil out.
Andy Brown is the Green Party councillor for Aire Valley in North Yorkshire