Will the AI growth continue?

Summary and recommendations

The eleventh evolution of artificial intelligence (AI) has taken the world to bed. Technology that only three years ago was considered complete science fiction has now become public domain.

For 250 kr a month whoever now have access to tools that can do whatever. Cheap AI models can now answer exams, create photorealistic images and videos, understand all the world's languages, write touching poems, compose songs, create complex code, etc. The list grows with each new update.

Today's powerful AI models are the result of a simple but powerful formula: the bigger, the better. More data, more computing power and more efficient algorithms give us more powerful AI.

If this scaling continues, we will get much more powerful models than we have today. The result is a society that will look radically different, where AI systems can perform all tasks humans can perform cheaper and better.

Will the scaling continue? In this paper, we examine three major barriers to AI development continuing as rapidly as today: lack of algorithmic improvement, data, and computing power.

We conclude that while there is a chance that AI development will slow down due to these barriers, it is more likely that it will not happen. The paper addresses:

• How scaling has created the current AI boom: How data, computing power and algorithms interrelate to create powerful AI systems.
• What can (but is unlikely to) stop development: Why familiar bottlenecks seem to be surmountable.
• What further scaling could mean for society: Why we must expect much more powerful AI going forward if scaling continues.
• What Norway can and should do: What policy proposals Norway should make in order to meet the development of AI.

‍

Policy proposals

In order for Norway to be better equipped for further developments in the field of ICT, we recommend that:

1) Norway establishes an AI Safety Institute. The first thing Norway should do is build sufficient expertise to follow developments closely so that we can be prepared for what is to come. Norway should therefore create an AI Safety Institute, as the UK and the US have already done, and Sweden is considering.

2) The development of the most powerful models must be regulated internationally. Norway should support international regulatory trials that set requirements for the largest models in the form of third-party evaluations, documentation, security procedures and access control.

3) The Oil Fund sets the AI standard. The oil fund must demand that the companies it invests in develop and use AI responsibly. It will be in the long-term self-interest of the fund and Norway.

4) Norway develops understandable and safe AI models. This needs to happen across academic communities and like-minded countries. We need a comprehensive European research collaboration; a “CERN” for safe AI.

5) We plan for different AI scenarios. AI development can stall, slow down or continue as it has done in recent years. The different scenarios produce radically different consequences for people and society. Both the public and private sectors in Norway should assess the consequences of and be prepared for all scenarios.

Download and read the full paper here: (NB: The PDF is in Norwegian.)