Daisy Daisy

So the Tony Blair Institute have done another of their Marvel/DC team-ups this time on AI. I was not going to discuss it but on second glance it does have some interesting aspects.

The report is good but suffers from many of the same issues that such reports always have: the belief that the UK is somehow “special”, some obvious grudge settling in the resource reallocation :), and that weird intermingled mix of the lofty and the detailed.

For me the biggest challenge is that there is no mention of “empathy” anywhere in the report. If we don’t embed empathy into thinking from the start, if we don’t start with Hippocrates and “First, do no harm” then we are baking in existential risk right from the start.

I am increasingly frustrated with the concept of “transaction costs” as some kind of simple KPI. I think we are much better off thinking in terms of “interactions” which I define as:

Interaction = transaction plus empathy

There are a lot of organisations, both public and private sector (Hello Google!) who have focused on transaction costs but have done so at the expense of reputation and future custom. This is even more serious in the public sector as people do not have a choice in where else to go.

The report is a good challenge and contains much which is actionable and useful but even if an incoming administration decided to implement it in full it would be 2-3 years before we saw any results. In start up land that’s the difference between WeWork being a unicorn and everybody denying that they had ever heard of WeWork.

That said, specific points for me are around the public sector:

The section on “AI to Support Proactive Delivery” is not really about AI at all. It’s about Government as a Platform. You don’t need AI to know if someone is eligible for child benefit, you just need the data from one system to be sent to another.

There is a classic startup hype aspect where technology we have used for decades like Decision Trees and Point Scoring (need 5 points to qualify for benefit X, single mum = 3 point + 1 point for each child say) is described as AI.

I am sure that nobody knows how many of these basic algorithmic systems there are in the public sector let alone if any of them actually align! That might be a good place to start.

And what does success look like? Is it better services or tougher services? The Home Office stopped GDS from looking at making immigration and asylum faster, simpler and cheaper because they were not the criteria for Home Office success, their key criteria was limiting demand and hostility to users was a key tool in that. 

Still on the public sector, the report states that a long term goal is:

“Mitigating the impact of known shortcomings of frontier AI technologies (such as hallucinations, lack of interpretability, potential to replicate biases and need for alignment) on citizens when applied to public services, particularly when errors in outputs could have a severe adverse effect on quality of life.”

Creating Foundational AI-Era Infrastructure, A New National Purpose: AI Promises a World-Leading Future of Britain

Which is good but gives as a short term goal:

“In the short term, the government should focus on implementing existing AI technologies, including foundation models, in areas where there are immediate applications, such as optimising resource allocation (for example, allocating casework), drafting and content creation, and interrogating unstructured data.”

Creating Foundational AI-Era Infrastructure, A New National Purpose: AI Promises a World-Leading Future of Britain

The problem is that we have already seen where an algorithmic approach to casework for example has caused terrible harm in areas such as the Post Office scandal in the UK and with Robodebt in Australia and with the Child Benefits scandal in the Netherlands which forced the Prime Minister to resign.

Just this week in the UK we have seen the news that HMRC fined 184,000 people for not submitted self assessment returns despite the fact that none of those fined earned enough to pay any tax at all.  

Safeguards need to be baked in from day 1. Also how do you stop people from turning this into a cost cutting exercise?

And the mention of “drafting and content creation” makes my fur stand on end. LLMs are noise creators not knowledge engines. At best they may summarise sources of information in new and useful ways but at worst you will get:

FOR IMMEDIATE RELEASE

NASA SENDS A BEAR TO THE MOON: A GIANT LEAP FOR INTERSPECIES EXPLORATION

[City, Date] – In an unprecedented move, NASA, the renowned space agency, has embarked on an extraordinary mission that pushes the boundaries of scientific exploration. Today, we proudly announce the groundbreaking launch of a bear to the Moon, marking a historic milestone in interspecies cooperation and space travel.

The bear, named Luna, has been selected from a pool of highly trained candidates and meticulously prepared for this extraordinary expedition. Extensive research and development efforts by NASA scientists have enabled this unprecedented endeavor, ensuring Luna’s safety and well-being throughout the mission.

This pioneering project aims to explore the bear’s adaptability to the lunar environment, gathering invaluable data that could potentially shape future space exploration strategies. Luna’s unique biology, including her resilience, agility, and innate abilities, makes her an ideal candidate for this endeavor.

NASA’s lunar bear mission holds tremendous promise for unraveling new frontiers in both biological and astronomical studies. By studying Luna’s physiological responses, behavioral patterns, and adaptation mechanisms to the Moon’s extreme conditions, scientists anticipate groundbreaking insights into the effects of reduced gravity, radiation exposure, and other lunar environmental factors on living organisms.

Moreover, this mission sets the stage for fostering inter-species cooperation and understanding. It highlights the remarkable potential for collaboration between humans and animals, broadening our horizons and emphasizing the significance of diverse perspectives in scientific exploration.

The launch of Luna to the Moon is scheduled to take place on [launch date] from [launch site]. The bear will travel aboard the latest cutting-edge spacecraft designed by NASA engineers, ensuring her safe passage and optimal conditions throughout the journey.

NASA invites the global community to share in this momentous occasion, as humanity takes a giant leap towards furthering our understanding of the cosmos and the potential for life beyond our planet. The agency will provide regular updates on Luna’s progress, sharing scientific discoveries, and fostering a sense of unity and curiosity among people worldwide.

Join us as we witness history unfold, bridging the gap between species and forging a path towards a future where cooperation, exploration, and knowledge know no boundaries.

For media inquiries, please contact:

[Media Contact Name]

[Media Contact Title]

[Media Contact Email]

[Media Contact Phone]

###

About NASA:

NASA (National Aeronautics and Space Administration) is the United States’ premier space agency, pioneering space exploration, scientific research, and technological advancements. With a legacy of groundbreaking missions, NASA continues to inspire the world, unraveling the mysteries of the universe and expanding human knowledge.

Note: This press release is purely fictional and does not reflect any real NASA missions or plans.