Progression in AI
As AI progresses faster than ever, one of the questions being asked by experts and the public alike is whether we can trust how decisions taken by intelligent machines are made. We’ve alluded in a previous blog to how robots are now packing groceries – a fairly benign task, sure – but as we’ve also touched upon before, there are claims that the NHS, among other sectors, could cut costs by making use of artificial intelligence.
The expansion of AI
Of course, AI requires extensive R&D software development, which naturally takes time to perfect, but this experimentation and development is becoming more and more prolific across a range of industries. From medicine to the military, it seems every sector wants to invest in AI.
But there are questions being asked – and, no doubt, solutions to be found – in the tech industry. What, for example, would happen if a machine supplied a patient with the wrong drug, or the wrong amount of that drug? Or if health insurance was refused to an applicant because of an algorithm?
One branch of research is focusing on creating neural networks – a vast number of interconnected processors that are intended to mimic the way the human brain works. These are able to handle a huge amount of data, identify patterns in it and adapt in response to learnings. While this means some exceptional leaps forward – such as improved accuracy in the identification of disease – it also means the machines become so complex that even their creators don’t understand the logic behind an answer that is given. This in turn means any flaw in the logic could be difficult to identify and correct, resulting in potentially unwanted outcomes.
The way forward
The industry is exploring ways to deal with the issues AI is presenting by experimenting with algorithms that can remain under human control, as well as trying to understand how machines reach specific decisions.
Companies are looking into the safe operation of artificial intelligence – that which can operate reliably and indiscriminately. Only then could a machine be put in charge of making a more important decision. Guaranteed higher accuracy than humans could achieve may be more important in a driverless car, for example, than our ability to understand how the machine reached a certain decision. On the other hand, knowing how a decision was reached is paramount if AI were to be used for criminal sentencing.
Either way, there is general agreement among professionals that safeguarding measures are needed to ensure any decisions being made can be accounted for.
AI development in the UK
If your business is engaged in the development of AI algorithms, you may be eligible to claim R&D tax credits. For details of how to calculate R&D tax credits or help with your claim, contact R&D Tax Solutions today.