AWS Executive: Only 8% of AI Models Reach Production
Experts from Amazon Web Services and McKinsey Quantum Black discussed practical challenges in deploying artificial intelligence. A small percentage of AI models in development are making it into production.
In a discussion hosted by Amazon Web Services (AWS), significant challenges in bringing artificial intelligence models to production were identified. Tom Godden from AWS and Matt Fitzpatrick, CEO of Invisible Technologies and former leader at McKinsey Quantum Black, shared insights into why only about 8% of AI models successfully move to actual use.
Fitzpatrick emphasized that AI adoption in enterprises is more complex than many anticipate. Unlike traditional software development, AI, particularly generative AI, requires continuous experimentation and learning. Companies often struggle with defining success metrics and ensuring the reliability and safety of models.
One of the major hurdles is generative AI's tendency to "hallucinate," producing incorrect or fabricated information. This, combined with the difficulty of managing organizational change, slows down deployment. While identifiable and manageable risks exist, many companies hesitate to initiate AI projects due to fear of negative publicity.
However, significant future opportunities were also identified during the discussion. Particularly, the increased efficiency in software development through generative AI is seen as revolutionary. Fitzpatrick estimates that within a few years, AI could significantly accelerate the development of new applications and boost engineers' productivity, opening new avenues for business and technology accessibility.