At a time of full employment and a bulging fiscal surplus, Minister for Transport and leader of the Green Party Éamon Ryan has asserted that the State now needs to employ thousands more civil servants. In an Inside Politics podcast for this newspaper last week, he claimed there was a clear understanding and agreement between the Government, employers and unions that the State’s public services need to be scaled up to match the success of the economy.
This year is seeing an extraordinary revolution in technology, in my view quite possibly as significant a transformation as the introduction of the printing press in the mid-15th century. The capabilities of large language models, as exemplified by ChatGPT (from OpenAI), Bing Chat (from Microsoft) and Bard (from Alphabet), suggest inflection opportunities for most business sectors to innovate new products, streamline operations and ease staff workloads.
Mundane tasks can become more automated but, as importantly, we can be encouraged and aided to engage our intuition and creativity throughout the workplace.
The competitive consequences of artificial general intelligence systems are being actively debated across the private sector worldwide. Could the new technology also deliver public services at an even higher standard, with nationwide consistency and seamless integration of multiple agencies and bodies, reducing the need for myriad new job positions across the Civil Service?
Some administrations have been trialling natural language chatbots for some time, not least our own Revenue Commissioners regarding tax clearance inquiries by telephone. One of the most comprehensive systems to date is the “Ask Jamie” service offered by the Singapore government, integrating more than 70 government websites with a philosophy of “no wrong door” for a citizen’s inquiry.
Why do some shareholders in the Republic's largest private residential landlord feel shortchanged?
These projects predate this year’s artificial general intelligence renaissance. With artificial general intelligence, the general public in principle should be able to freely use ordinary language to access information about all government services to find guidance and advice. Likewise, public service staff should now also be able to use natural language to interact with internal systems, whether querying databases, analysing data or tracking fraud. Legislation, policy papers, reports and correspondence can all be better informed and researched, and all can be rapidly and accurately summarised as briefs as required.
Also last week, the European Council’s internal analysis and research team published a white paper on ChatGPT in the Public Sector. The report reviews opportunities and risk mitigation across the eight European principles of public administration, including accountability, transparency, impartiality and reliability.
Large language systems require an expensive brute force approach to their training and one that only a few technology companies can afford
A Green Party minister might object to the carbon footprint involved in training large language models. However, the figures given by OpenAI for the training of its ChatGPT show it required less than a quarter of the total power consumed annually by the data centres wooed into Ireland by the Government.
More significantly, artificial general intelligence systems are trained using the open internet and thus need extension to answer queries relating to confidential data in the public sector, such as urban planning, welfare or taxation.
The private sector of course faces a similar requirement for queries over commercially sensitive data. Software developer forums are flourishing with how to fine tune artificial general intelligence (AGI) systems and providing contexts for queries into proprietary data. PwC has very recently announced a substantial commitment to Microsoft’s AGI technology, which includes guarantees that PwC commercial information and client data cannot be inappropriately accessed, including of course by Microsoft itself.
Large language systems require an expensive brute force approach to their training and one that only a few technology companies can afford, currently all of them American. The resultant concentration of power in these companies and the proprietary nature of their offerings means that as yet there is no independent oversight to inspect and analyse their models.
Rather than being trained by just analysing and predicting text patterns, the next versions of AI may be augmented by knowledge widely accepted as common sense
However, the main political groups of the European Parliament have just last week reached agreement on the AI Act, now expected as a result to pass a plenary vote in June. The draft Act includes timely provisions relating to artificial general intelligence, including high-risk uses, safeguards against bias (for example in religious views or sexual orientation) and disclosure obligations of copyrighted information in the training of machine models.
The technology is also rapidly evolving. Within the artificial general intelligence community, there is a growing acceptance not to further expand the extreme scale of large language models, but instead to move towards even better performing smaller systems. Rather than being trained by just analysing and predicting text patterns, the next versions may be augmented by knowledge which is widely accepted as common sense by humans. Sam Altman, the OpenAI chief executive, hinted last week that artificial general intelligence systems can be improved in other ways than by simply making them even larger.
[ AI speed cameras capable of detecting phone and seatbelt use considered by GardaOpens in new window ]
The State is apparently considering a long-term investment and financial commitment to the expansion of the public service. I urge that the rapid advances in technology would be a major consideration in how to best provide a transparent, consistent and socially just society.
