Why Generic AI Solutions Fail
General-purpose AI platform integrations often end up as expensive research projects.
The most effective AI solutions learn from the historical data and provide relevant, actionable, and highly accurate insights to solve a business problem. One of the core challenges with these solutions today comes from understanding the domain-specific intents hidden inside proprietary operational data. (“Intent” – what the data means, and what it might require a person to do. Context.)
Extracting usable intent can be onerous – and the difficulties involved can easily drive costs beyond the point of positive ROI. Low accuracy and the inability to adapt to domain specific problems neuter the efficacy of general-purpose AI algorithms and platforms.
For example, natural language processing techniques like the Stanford NLP libraries are built from the English language to discover part of speech like noun phrases in an existing conversation. The problem is that the domain specific conversation used by technicians, physicians and other specialized experts doesn’t sound like the English language or closely follow its grammatical patterns. The structures are dissimilar enough that generic AI cannot provide anything near actionable insights.
The other challenge with existing AI platforms is that they provide tools for other data scientists, and are far from readily usable for business users.
The answer to these challenges comes in specialized ‘blackbox’ AI platforms and solutions built around a specific business use case, with the configurability and adaptability to be quickly and easily applied to new enterprises and industries. For example: the repair and maintenance of complex equipment. Airplanes and commercial dishwashers are completely different, but the repair technicians that service them could sit down over a beer and share war stories.
This balance — specialization to a specific problem set, juxtaposed with flexibility to the proprietary formats and details — is the key to next-gen artificial intelligence. Why is it necessary? A good example is how the over-aggressive ambition to be the digital twin operating system for an entire industry resulted instead in the entire department getting fired — GE Digital’s ‘pivot’ earlier this year.
Businesses should start their AI powered transformation journey by looking for use-case specific solutions that bring in faster ROI with quick proof of value initiatives. “We don’t know what we don’t know” is a dangerous place to start. Businesses should also revisit their data ecosystem and make adjustments in data collections to focus more on strategic value sources vs pure operational data collection.