Reflections on the links between Living With Artificial Intelligence BBC Reith Lectures and Living With Data
By Monika Fratczak, new Living With Data team member
Professor Stuart Russell delivered four fascinating BBC Reith Lecturer lectures in 2021, which explored the impact of artificial intelligence, data-driven systems, automation and machine learning on people’s lives. Russell argued that the world’s approach towards and increasing reliance on data-driven and algorithmic systems that are not guaranteed to be error-free represents a profound historical change. He predicted that we could lose control over super-powerful AI, with potential risks for the future of human lives, if we don’t address the question of how to regulate AI systems to make sure that they are human-centric and “they do not produce disastrous outcomes.”
Russell argued that, as individuals,we have no power and control over the information technology industry or over our data, which are collected analysed, shared and used by “corporate representatives sitting in [our] pocket busily sucking out as much money and knowledge and data as they can.” In his opinion, manipulative social media algorithms provide a major threat, because manipulating human preferences changes what people want. He claimed that AI experts “don’t even have a basic philosophical understanding of how to make decisions on behalf of someone who’s going to be different when those decisions have impact.” It’s important to call out the power of global technology companies as Russell does, but at the same time, Living With Data findings suggest that he underestimates people’s understanding of them and people’s agency when engaging with data and AI.
Across all four lectures, Russell returned repeatedly to the question of human control over AI systems, arguing that the world’s current approach to AI needs to change to “ensure machines do the right thing.” Despite highlighting many negative consequences of AI – including the immorality of creating algorithms that can decide to kill humans; racial and gender bias; spreading disinformation and deepfakes; and cybercrime – Russell optimistically concluded that some tech companies are genuinely interested in how their products can be more beneficial to people. He said that he and his teams are “developing research relationships and we’re finding openness and willingness to share data and algorithms so that we can actually understand how to do this right”.
Russell claimed that it is technologically feasible to design AI systems that are beneficial to people and represent users’ interests. He proposed a new model based on three principles:
- the machine’s only objective is to maximise the realisation of human preferences;
- the machine is initially uncertain about what those preferences are;
- the ultimate source of information about human preferences is human behaviour.
This radically different approach to AI design aligns with our aims on Living With Data, as well as the aims of bodies like the UK government’s Centre for Data Ethics and Innovation (CDEI), which wants to ensure that data and AI are ‘a force for good’). Living With Data’s findings could help to realise these aims, but it’s important to ask whether realising human preferences is the most pro-social aim of AI implementation, and whether human behaviour, rather than feelings and experiences, should inform future action.
Listen to the four BBC Living With Artificial Intelligence Reith Lectures by Stuart Russell here: