AI fears serious, but frontier slowdown unlikely: BMO

In a Friday note, BMO noted that concerns over artificial intelligence safety were justified, though leading AI labs displayed no signs of slowing down their development efforts. BMO's AI specialist stressed that fears among AI leaders were legitimate, given the critical importance of aligning AI systems with human values before issues escalated.

However, the brokerage questioned the commitment of developers like Anthropic and OpenAI to pace their progress, citing the release of increasingly powerful models. BMO pointed out that recent AI-related hacking incidents were relatively minor, yet warned that malicious actors could exploit open-access models more easily. Users had demonstrated the ability to circumvent safeguards, prompting models to generate misaligned results.

Concerns over AI safety escalated after Anthropic CEO Dario Amodei called for a coordinated slowdown in frontier development, citing potential risks to humanity. Some AI executives agreed with his proposal, and discussions on AI safety were held during a recent U.S.-China summit. BMO anticipates both the U.S. and China to eventually implement stricter AI regulations, focusing on open-source and open-access models rather than closed systems.

Such regulations would likely require approval before the public release of open-access models, BMO explained. Both governments have an incentive to restrict AI openness, which would strengthen pricing power for closed-source frontier labs like Anthropic and OpenAI. BMO expects token prices to continue falling, increasing the usage of AI tools.

However, prices are likely to hit a bottom around 2030, after which closed-source providers will likely raise prices as users become dependent on their tools.

Meta Platforms' AI strategy improved due to better utilization of its AI capabilities, but the company still trails behind frontier leaders such as Anthropic and OpenAI. BMO remains bullish on AI hardware, citing the scarcity of computing power as users realize that AI's capabilities have surpassed earlier expectations. BMO indicated that white-collar job displacement would become significant by 2028, while blue-collar jobs would be affected later, in the 2030s.

This later impact was attributed to Moravec’s Paradox, which suggests that tasks requiring physical dexterity are more challenging for AI to automate compared to cognitive tasks.