Artificial Intelligence is an ever-evolving field, with its rapid pace of development often leaving us intrigued and, at times, skeptical. In the midst of this AI whirlwind, DeepMind, Google’s subsidiary AI research and development laboratory, has unveiled a paper outlining a framework for assessing the societal and ethical implications of AI systems. This revelation comes at a pivotal moment as global discussions on AI ethics gain momentum.
The paper introduces a compelling perspective, emphasizing the need for a multidimensional approach to evaluate and audit AI systems. It suggests that AI developers, app developers, and a broader spectrum of public stakeholders should contribute their insights to this critical process.
This unveiling aligns with the forthcoming AI Safety Summit, which the U.K. government is hosting. The summit is poised to assemble international governments, leading AI companies, civil society organizations, and research experts. The primary focus will be on effectively managing the risks associated with recent advancements in AI technology. One of the key initiatives to be introduced at the summit is a global advisory group on AI, reminiscent of the Intergovernmental Panel on Climate Change, designed to provide regular reports on the cutting-edge developments in AI and the associated risks.
DeepMind’s proactive approach to sharing its perspective with the world before the summit highlights the importance it places on contributing to the policy discussions that will shape the future of AI. Within this perspective, DeepMind suggests examining AI systems at the “point of human interaction” and considering the ways in which these systems become integrated into society.
Yet, while DeepMind’s proposals may be promising, it is essential to scrutinize the transparency of the lab and its parent company, Google. Recent findings by Stanford researchers, who ranked major AI models based on their transparency, suggest that Google’s flagship text-analyzing AI model, PaLM 2, only scored 40% on a transparency index, raising concerns about the company’s commitment to openness.
DeepMind, although not directly responsible for PaLM 2, has also faced challenges related to transparency in the past. These findings shed light on the need for more transparency across the AI field, with top-down pressure being a vital driver for improved practices.
However, DeepMind seems to be taking steps to address these transparency concerns. Collaborating with OpenAI and Anthropic, DeepMind has committed to providing early or priority access to its AI models to support research on evaluation and safety, particularly for the U.K. government.
The AI community is now looking forward to DeepMind’s forthcoming AI chatbot, Gemini. The company’s CEO, Demis Hassabis, has described Gemini as a rival to OpenAI’s ChatGPT. To establish credibility in the realm of AI ethics, DeepMind will need to comprehensively outline both the strengths and limitations of Gemini.
DeepMind’s proactive approach to sharing its perspective on AI ethics indicates its commitment to the field. However, the forthcoming months will be a crucial test of its ethical aspirations. Ensuring transparency and open dialogue about AI’s societal and ethical implications is a responsibility that extends beyond a single lab or company. It is a