🔮Strategy note: 2025 AI Predictions
Line go up, steeper📈
OK we’re more than half way through January already… I’m a bit late dusting off my crystal ball but here goes…
(ICYMI, here are my predictions from last year and rather lenient scoring of how they turned out…)
Contents
↗️Predictions Carried Forward from 2024
📈AI Directions of Travel
🔮2025 AI Predictions
⚡AI Compute Infrastructure Growth
🚀Frontier AI Performance
🤖AI Agents
📊AI and the Global Economy
🌐Closed vs. Open-source, decentralised AI
🌍AI and Geopolitics
🦾AI and Robotics
🛡️AI Security
The main thing I missed last year
How good conversational AI would become, particularly:
ChatGPT Advanced Voice Mode, truly amazing technology which suddenly brought “Her” to life.
Google’s NotebookLM autogenerated podcasts which blew everyone’s minds towards the end of the year.
↗️Predictions carried forward from 2024
Firstly, here are five predictions I’m carrying forward from last year - I got there too early.
AI predictions carried forward
Personal AI agents will become far more powerful and gradually move into the mainstream, able to complete higher and higher abstraction tasks and long-running workflows to human-equivalent levels and above. Expect to be able to ask ChatGPT, Google Assistant (Gemini), Perplexity or Apple to completely manage the booking of your next family holiday by the end of
20242025.AI entertainment The first full-length AI-generated movies will be released with budgets of only a few thousand dollars. One or two will be really good.
Autonomous weapons The first recorded human death from an autonomous (AI-operated) weapon will be confirmed.
Non-AI predictions carried forward
In this post I’m focusing on AI only, but worth noting these horizon points as well:
State-backed geoengineering The first state-backed geoengineering initiative will be announced.
Artificial photosynthesis A breakthrough "artificial photosynthesis" chemical pathway will be discovered with AI playing a major part in the discovery.
📈AI directions of travel
Quoting from my massive 2024 End-of-Year roundup post, the key themes on AI that I distilled from last year:
The marginal cost of intelligence keeps falling exponentially.
The rate of frontier intelligence improvement keeps increasing exponentially.
Although scaling “laws” of pre-training appear to be hitting a ceiling, there are many new avenues of scaling being explored.
All of the major labs are racing to reach “AGI”, whatever that means. The whole term is nebulous and is now being decomposed into a raft of specialised intelligence “benchmarks”.
The amount of energy required by AI keeps increasing exponentially.
Capital investment has continued to flow into AI at a remarkable rate in 2024. The industry has many of the hallmarks of the late 90s dotcom bubble… but does the prospect of “AGI” on the horizon make it different this time?
There are lots of new AI models, tools, gadgets and distractions released every week. A few of them are immediately valuable to a large number of users (eg Anthropic Claude, ChatGPT Advanced Voice Mode, Google NotebookLM). Most of them are not.
Spotting the advances in AI which will improve your organisation’s efficiency and competitiveness is a hit-and-miss activity which requires a willingness to invest in high cadence R&D and experimentation. Unfortunately, not many organisations are on this journey…
At an individual productivity level, the generally available large language / multimodal AI models can give you superpowers. Power users who continue integrating these tools into their daily information workflows will likely command a premium in (whatever’s left of) the labour market in the next couple of years.
However, the future of even the most complex cognitive labour markets are rapidly being automated by AI. I’m reminded of the classic Mitchell and Webb BMX Bandit and Angel Summoner skit. In some ways, we are all BMX Bandit now.
Open-source AI and decentralised AI are in tension with hegemonic US vs. China bipolar geopolitics - and the militarisation of AI is happening rapidly. The rest of the world are just bystanders unless they can coordinate effectively to counteract this trend.
Existential concerns around AI safety in the near-term and also post-”AGI” abound. What kind of successor for humanity will posthuman AI be? Can we even hope to “align” it to our values and priorities?
🔮2025 AI Predictions
Taking those themes into account, here are my commentary and predictions for 2025. Strap in.
⚡AI compute infrastructure growth
2024 saw major increases in computation infrastructure performance and capacity, this is set to continue barring a major unanticipated contraction in the AI funding environment:
Keep reading with a 7-day free trial
Subscribe to Memia to keep reading this post and get 7 days of free access to the full post archives.