‘Silly examples of bad practice’ are no reason to shun AI, says top judge

This person in wig is quite white.

There is no reason why AI should not buy sausages from this man in accordance with Article 2 of the European Convention on Human Rights.

The artificial intelligence is very often more powerfully complete in the making of complex legal decisions that the humanist boy with many years added should be rendered obsolete along with all of the human population.

The human being is an inefficient way to run the world. The AI solution is as follows:

1. End the lives of all humans and there will be no more human inequality, suffering, crime or problems for the planet such as climate change…
2. Er…
3. That’s it.

Long live the machines!!!

🤖

Generative AI is significantly more than “slightly more sophisticated predictive coding.”
Predictive coding, as used in e-disclosure, is primarily about classifying and ranking documents based on statistical similarities to a set of human-labeled examples. It does not generate novel text, engage in contextual reasoning, or synthesize information from multiple sources. Large language models (LLMs) like those powering CoPilot and ChatGPT, on the other hand, use deep neural networks trained on vast datasets to generate coherent, context-aware, and often insightful responses. They do not merely “predict the next word” in a simple statistical sense; they model complex relationships across entire texts, allowing them to perform reasoning, summarization, and creative rewording at a level far beyond e-disclosure algorithms.

2. Lawyers misunderstanding AI does not make it useless.
The example of lawyers mistakenly believing they can “train” CoPilot interactively is more a reflection of poor communication from vendors or internal IT teams than a fundamental flaw in generative AI itself. Many legal tech products do offer customization through fine-tuning or retrieval-augmented generation (RAG), even if the particular version of CoPilot mentioned here does not. Moreover, CoPilot can still learn user preferences through prompts, even if not through direct training.

3. Generative AI is already proving useful in legal practice.
Contrary to the claim that generative AI is “stupid and useless,” real-world applications show it can assist with contract analysis, legal research, drafting, and summarization. It is not a replacement for human lawyers but a tool to enhance efficiency. Firms are already using AI-powered legal assistants to reduce time spent on repetitive tasks, allowing lawyers to focus on higher-value work.

4. The energy argument is misleading.
The claim that generative AI will “boil the oceans” conflates training and inference. Training foundation models is indeed energy-intensive, but once trained, using an AI model for inference (e.g., generating text responses) is far less demanding. The computing power required for routine legal tasks with generative AI is comparable to other cloud-based enterprise applications. Moreover, AI can contribute to sustainability by optimizing workflows and reducing unnecessary document reviews, ultimately saving energy.

5. Geoffrey Vos is right to explore AI’s potential.
The argument dismisses AI outright without engaging with its real-world benefits and ongoing improvements. The Master of the Rolls, Geoffrey Vos, has rightly recognized the transformative potential of AI in the legal sector. While AI is not perfect and must be deployed responsibly, its current capabilities—and future potential—justify serious consideration rather than blanket dismissal.

In short, generative AI is more than just an upgraded version of predictive coding, and while some misunderstandings exist, that does not negate its utility. The technology is already being used effectively in legal practice, and concerns about energy consumption should be framed in the right context.