Is it possible to create ethical AI?
Emma Kendrew, artificial intelligence lead at Accenture Technology UKI, says YES.
Yes, we can create ethical artificial intelligence (AI). But you must acknowledge that you cant create AI one day, and have a fully matured ethical framework the next.
AI needs to be nurtured in the same way as a child. It needs to be educated and taught the principles of good citizenship: responsibility, fairness, and transparency.
The key is to give AI the right teacher (diverse teams representative of society) to design and build it, the right information (unbiased and verified data) to learn from, and the right corrective process (ongoing training and education). Lets also remember that, alongside learning, there is the need for “unlearning”, if it falls in with the wrong crowd (e.g. biased data).
By combining these principles, we can foster an understanding of right and wrong, and what it means to behave responsibly. For some businesses that will mean big process changes, but it can and should be done.
Alastair Benn, editor of Reaction Weekend, says NO.
Science fiction is full of omnipotent, hyper logical super-computers, like Alpha 60 in Jean-Luc Godards 1965 film Alphaville. Alpha 60 is actually pretty keen on ethics. He informs us: “rest assured that my decisions always have in view the ultimate good”.
Artificial intelligence (AI) does preserve a notion of the “good”, but its an essentially utilitarian mode of thinking. Utilitarianism emerged from continental traditions of positivism in the early nineteenth century. For thinkers like Helvetius and Saint-Simon, the proper experience of ethics could only emerge under a centralised political order, with a high degree of social planning.
Its a vision thats quite alien to radical British traditions of political liberty. At around the same time, in the writing of Locke, and Stuart Mill, ethics emerged as an imperfect art that accommodates between the conflicting projects of flawed individuals, without the interference of the state or higher intelligence.
AI could have some form of ethics hard-wired into its mainframe – its just not a particularly desirable one.