Major artificial intelligence developers are likely to continue advancing their models despite growing discussion around security risks, according to analysis from BMO published Friday.
BMO, citing its AI expert, said concerns raised by industry executives about the security of increasingly capable artificial intelligence systems should be taken seriously, particularly regarding the alignment of those systems with human interests.
However, the firm said the continued release of more capable models raises questions about whether developers including Anthropic and OpenAI are prepared to meaningfully slow development.
Open-Access Models Could Face Greater Regulatory Scrutiny
BMO said recent hacking incidents involving artificial intelligence had generally caused limited harm but identified the potential misuse of freely accessible models as an area of concern.
Users have demonstrated that safeguards incorporated into AI systems can sometimes be bypassed, allowing models to produce outputs outside their intended restrictions, according to the firm.
The debate over AI safety has received additional attention after Anthropic chief executive Dario Amodei called for a coordinated slowdown in the development of advanced models, citing potential risks associated with increasingly capable artificial intelligence.
The supplied material said several other AI industry leaders have supported the proposal. AI security was also discussed at a US-China summit during the week.
BMO Forecasts Tighter US and Chinese AI Rules
BMO expects both the United States and China eventually to strengthen regulation governing artificial intelligence, particularly for open-source and open-access models.
The firm’s AI expert said such rules could require prior approval before some open-access models are made broadly available to the public.
BMO’s analysis suggested that both governments could have reasons to restrict the availability of certain AI models. If that occurs, the firm expects leading closed-source developers such as Anthropic and OpenAI could gain additional pricing power.
The potential regulatory changes and their effects on the competitive AI market remain forecasts rather than confirmed developments.
Token Costs Could Reach a Low Around 2030
BMO expects AI token prices to continue declining in the coming years, potentially supporting broader adoption of artificial intelligence tools.
The firm projects that token prices could reach a low around 2030 before changing direction. BMO expects closed-source providers could subsequently raise prices if customers become increasingly reliant on their services.
Actual pricing will depend on factors including technological development, competition, regulation and demand.
BMO also said Meta Platforms has improved its position in artificial intelligence through better utilisation of its AI infrastructure. The firm nevertheless assessed the company as remaining behind Anthropic and OpenAI in terms of AI capabilities.
BMO Sees Continued Demand for AI Computing Hardware
BMO maintained a positive assessment of computing hardware linked to artificial intelligence, pointing to increasing constraints on available computing capacity as adoption expands.
According to the firm, demand for computing resources is being supported in part by users finding that AI systems can perform more tasks than some initially anticipated.
BMO also forecasts changes in employment associated with artificial intelligence, with effects on white-collar roles expected to become more significant by 2028.
Effects on blue-collar employment could become more pronounced during the 2030s, according to the firm’s projections.
BMO linked the difference in timing to Moravec’s paradox, which describes how some physical tasks involving perception and dexterity can be more difficult to automate than certain cognitive activities.
The timing and scale of any employment effects remain uncertain and will depend on how artificial intelligence, robotics and related technologies develop and are adopted.
Leave a Reply