The article also mentions the sharp edge formation of AI capabilities, where AI exhibits capabilities surpassing humans in certain tasks while showing limitations in other seemingly simple tasks.
- Neglect of AI Future Planning and Reasons
Phenomenon of planning neglect: Although planning cycles often span more than a decade, few organizations seriously consider the possibility of continuous AI improvement in their strategic planning.
Divergent views: AI researchers have differing opinions; some believe that capabilities will grow exponentially in the foreseeable future, while others think that large language models have reached their limits. Regarding AGI, some insiders believe it could be achieved within 3-5 years, while other researchers think it will take longer. A 2023 survey of computer scientists shows an average expectation of 2047, with a 10% chance of achieving it by 2027, and the prediction market suggests 2033.
Capability forecast: The author believes it is highly likely that by 2027, models will be able to perform the work of AI researchers/engineers.
Overemphasis on superintelligence: Future-oriented AI discussions often focus on superintelligence, which is too distant for most people and leads to a lack of planning. However, even without considering further development of AI, current AI systems poorly integrated into businesses and organizations have already caused many problems, such as affecting employment, triggering phishing, and changing educational methods.
The incentive role of AGI: AGI serves as an incentive goal for the industry. Even if it cannot be achieved, investment will continue to enhance AI capabilities, and the current integration of AI into life and work will bring about transformation.
Lack of documentation: Major AI laboratories lack in-depth documentation for non-professionals, often providing vague suggestions and trial-and-error approaches, leading to an underestimation of AI capabilities. Many people have only used outdated AI and do not understand the capabilities of current systems.
Hidden functionalities: Features like Claude's artifacts and ChatGPT's Code Interpreter are often hidden and not easily accessible.
Blurred boundaries: Over the past two years, the boundary between human and AI capabilities has become increasingly blurred, with machines demonstrating creativity, empathy, etc.
Uneven capabilities: AI capabilities are uneven, performing well in some tasks and poorly in others. People often focus on their shortcomings, such as Claude's ability to generate application ideas and prototypes but not being able to build complete products.
- Suggestions for Planning the Future of AI
Multiple future possibilities: The author proposes four potential futures for AI, including a plateau of capabilities, linear growth, exponential growth, and AGI.
Planning method: Organizations should use scenario planning to consider different futures, with GPT-4o assisting in the process. In summary, we should stop ignoring the changes in the world and actively plan to take control of the future.