Dave Levy continues our discussion on the merits and threats of AI
There can be little doubt that the introduction of AI will be used to attack the terms and conditions of their white collar users.
What will be the macro effect of the wide-scale introduction of AI? Will jobs be lost? Many suggest so, but it is not certain. However, the casualisation of work that was once professionally skilled, particularly in white collar work, will be a contributing factor.
On jobs, it’s interesting to study a paradox for the 19th century, where the invention and rollout of railways in the UK increased the amount of horse-drawn transport. It’s possible that AI will extend the demand for a number of human services. There are others who argue, using US legal services as their example, that an increase in productivity may lead to lower costs, thus leading to poorer people having access to said legal services. This would be a good thing as it would give access to the courts and justice to the less well off, but it must be free, and the legal aid system must be repaired in the UK.
In the short and maybe medium term, one of the most important defences that we as citizens, workers, and consumers have is the EU’s GDPR, which in Article 22 & Recital 71 establishes what they call a right to “freedom from profiling”. This, through the rulings of the CJEU, has become quite extensive and now prohibits such things as ‘general monitoring’, a legal protection brought forcefully to light by the French supervisory authority fining Amazon €32m for violations of the GDPR within their workforce management regime.
Authority vs Popularity
Large Language Modules (LLMs) have been described as exceptionally expensive auto-complete machines. Because of this, an area of concern about their use relates to a question of authority in the answers. All the search engines have had their implementations refactored to present the search query answers in an enhanced human-readable form by AI, and bit by bit they are exposing their sources. However, when making some queries and decisions, the information required must be authoritative and not a popular opinion. Both the previous and current search ranking algorithms use popularity as a key factor in determining the query response’s sort order. This is not always wanted when the authoritative current policy is required. This is important when dealing with science or law.
Privacy and competitive advantage
An adjacent issue is where the knowledge base needs to be constrained or verified as true. When using third-party AI engines, your question and answer become part of the future and query authors may wish to keep this secret. Also, citizens should not want either the police or the courts taking their legal briefings from a public/popular rule/text-based. These and similar organisations need to build their own systems with a private (or publicly shareable rule-based, but one where the entry of data into the system is controlled to ensure the data is true and relevant to decision-making. Also, some private or third sector organisations will want to keep their research secret for reasons of competitive advantage and much data is covered by the GDPR and its “right to be forgotten”. There is no question that using a third-party, global LLM means your questions and answers contribute to the next set of questions and answers. In terms of intellectual property, you don’t want to be contributing trade secrets to the third-party LLMs, nor do you wish to have published personal data.
Next generation regulation
The attempt to create a global rules-based system would seem to be an attempt to broaden the datenkraken’s reach and influence. The EU’s Digital Services Act is a regulatory tool that allows the EU to regulate the top internet service providers and fine them if they breach EU law, which, in the case of the GDPR, is extra-territorial. Some EU NGO’s are engaged in using this law to create take-downs of hate speech, noting that in Europe and even in Brexit Britain, hate speech, is a criminal act. However, the UK’s legal system is ill-equipped to find such material, let alone take it down and punish its authors and carriers.
It’s interesting that while the US federal government is seeking to deregulate the IT giants, many states are adopting new more restrictive laws which leaves Sunak’s attempt, inherited by Labour, to offer a median regulatory regime in tatters.
In summary, there are plenty of laws to ensure that AI and its owners behave decently, and in some European countries, the will and resources to enforce them, but it’s not universal. Also, there are important economic countervailing forces opposing the creation of a privately owned “Global Intellect” even if the current technology is capable of such a task.
https://www.ft.com/content/aed82f47-b441-4bb3-930e-eca10585fc6d