Industry-focused review instead recommends ‘data trusts’ to establish framework for the tech
The suggestion was made by computer scientist Wendy Hall and BenevolentTech CEO Jérôme Pesenti in a review commissioned by the government in a bid to boost the AI industry in the UK.
The resulting report is strongly geared towards how government can incentivise the use of AI, with recommendations for increasing data-sharing and improving skills.
However, it stops short of proposing AI regulation, instead recommending the creation of an “AI council” as an oversight body and “data trusts” to establish a framework for the technology.
Which is in stark contrast to recent interventions from industry bods, such as Elon Musk’s call for proactive, not reactive, regulation.
Hall and Pesenti acknowledge what some will see as a gap in their work, but argue that “resolving ethical and societal questions is beyond the scope and the expertise of this industry-focused review”.
They instead point to a June report by the Royal Society and British Academy about governance of data, which called for a stewardship body to take a “helicopter view” of the field.
The pair said that the proposed overarching principles for data governance also be applied to AI, and that the stewardship function includes AI expertise.
“While AI will generate some specific challenges, it would not be helpful to see AI governance as something unrelated and separate to broader data governance,” the authors wrote.
Elsewhere in the report, Hall and Pesenti call for government to establish an AI council as a “strategic oversight group” to encourage an “open and non-competitive forum” to coordinate collaboration between industry, the public sector and academia.
The council would be an “expert leadership group” that offers advice to policymakers and is also responsible for identifying and tackling skills deficits.
A further recommendation is for the creation of “data trusts” between organisations holding data and those who want to use it to develop AI.
The idea is to provide a “repeatable framework” for data-sharing terms and conditions in a bid to standardise the process, which the authors said sees decisions made on a case-by-case basis.
An overarching support organisation for these data trusts should develop “tools, templates and guidance” to help organisations share and use data.
Other recommendations in the report include proposals to establish common policies and practices for licensing of IP and forming spin-out companies, and that the underlying data from publicly funded research be published in machine-readable formats.
Further recommendations on boosting skills include “breaking down stereotypes” around AI, creating 200 more PhD places dedicated to AI, and offering AI courses to MSc students so they develop more specialist knowledge. ®
