The Future Cost of IBM Watson-like Computers: When Will They Be Affordable?

When discussing the cost of powerful artificial intelligence systems, such as IBM Watson, it is important to consider not just the initial hardware cost but also the ongoing operational expenses. Traditional approaches often measure the cost based on the system itself, but a more practical metric is the cost per result or answer provided by the AI. This article will explore how advances in technology and software efficiency will bring down the cost of these systems to make them more accessible.

Current Costs and Expectations

According to some calculations, today it costs IBM Watson approximately $2.84 to answer a question when running at full capacity on their own system. However, with advancements in cloud computing, this cost can be significantly reduced. Using Amazon's cloud platform, the cost drops to around $1.50 per question. For those looking to build their own system, the cost is even lower, estimated to be around $0.50 per question before any software and hardware optimizations.

For perspective, Google's revenue per search query is approximately $0.20. This comparison highlights the potential cost-effective nature of advanced AI systems.

Predictions and Skepticism

Michal's points make a lot of sense. However, I remain skeptical about making precise predictions that far into the future. Instead, focusing on the potential for software improvements to reduce computational requirements might be more realistic and actionable. Clever software can often compensate for the limitations of a hardware platform, making it feasible to create a version of Watson that offers similar functionality with less demand on the computational hardware.

Trend Analysis and Future Predictions

The trend in computing power per dollar shows a stable doubling every 14 months. Hardware vendors predict that this doubling trend will continue for at least the next two decades. If we consider the initial cost of Watson hardware in 2011 at approximately $1 million, it is likely that the cost will drop below $1,000 by around 2023.

However, it is important to note that software costs are also a significant factor. In 12 years, the current software is very likely to become obsolete and need to be replaced with more sophisticated and computation-intensive alternatives. This software cost will add to the overall expense of maintaining an IBM Watson-like system.

Concluding Thoughts

While the future cost reductions for Watson-like systems look promising, it is crucial to approach these predictions with a balanced perspective. The continuous advancements in hardware and software efficiency can significantly reduce the cost of implementing and maintaining these systems, making them accessible to a broader audience.

By focusing on the cost per result rather than the initial hardware cost, it becomes clear that the path to affordable advanced AI systems is within reach. However, it is equally important to consider the ongoing software and hardware updates to ensure that these systems remain effective and economically viable over time.