Examining DeepMind’s Ethical Reliability

November 6, 2023

This is a Summary of:

"This week in AI: Can we trust DeepMind to be ethical?"

by Kyle Wiggers and Devin Coldewey

Published on

TechCrunch

Deep Mind Logo and Text

Key Takeaways

  • DeepMind, the Google-owned AI R&D lab, released a paper proposing a framework for evaluating AI systems' societal and ethical risks.
  • The lab calls for involvement from AI developers, app developers, and broader public stakeholders in evaluating and auditing AI.
  • DeepMind's parent company, Google, scored poorly in a study on transparency, suggesting a lack of pressure for DeepMind to improve.
  • DeepMind has committed to providing early access to its AI models for research into evaluation and safety.The lab's forthcoming AI chatbot, Gemini, will test DeepMind's commitment to transparency and ethics.

Summary of "This week in AI: Can we trust DeepMind to be ethical?"

DeepMind, the Google-owned AI R&D lab, has released a paper proposing a framework for evaluating AI systems' societal and ethical risks. The timing of the paper is different, as it comes just before the AI Safety Summit, where international governments, leading AI companies, civil society groups, and experts will discuss managing risks from recent advances in AI. DeepMind's proposals call for involvement from AI developers, app developers, and broader public stakeholders in evaluating and auditing AI.

However, the lab's proposals need to be more credible when looking at DeepMind's parent company, Google. A recent study by Stanford researchers ranked ten major AI models on how openly they operate, and Google's flagship text-analyzing AI model, PaLM 2, scored poorly. That suggests there is little pressure from the top for DeepMind to be transparent about its models.

Despite this, DeepMind has committed to providing early access to its AI models for research into evaluation and safety. However, it is still being determined if this is merely performative, as DeepMind is not known for philanthropy and generates significant revenue through licensing its work to Google teams.

DeepMind's forthcoming AI chatbot, Gemini, will significantly test the lab's commitment to transparency and ethics. CEO Demis Hassabis has promised that Gemini will rival OpenAI's ChatGPT in its capabilities. Still, for DeepMind to be taken seriously in the AI ethics realm, it must thoroughly detail the chatbot's weaknesses and limitations.

In addition to DeepMind's developments, other notable AI stories this week include flaws found in Microsoft's GPT-4, OpenAI's introduction of web searching and DALL-E 3 to ChatGPT, and the emergence of alternative models to OpenAI's GPT-4V. There are also stories about AI playing Pokémon, Google's AI-powered language tutor, Amazon's testing of warehouse robots, and China's investments in an OpenAI rival.

Click below to read the full story