Alert: Snake oil sellers now selling AI
There are plenty of different AI-based solutions on the market. However, only a few of these are based on human-centric cognitive AI that understands humans thus enabling automation. The problem is that there are many AI suppliers and almost all of them sell “true” AI that falls short on many essential capabilities. It is very difficult for customers to evaluate the capability of AI before seeing the results of a failed project.
This blog will answer the following questions: What is a cognitive human-centric AI? What are the benefits of a truly cognitive AI for business? How can the competence of an AI provider be verified?
Human-centric cognitive AI in a nutshell
Cognitive AI can make sense of a written text in a similar manner to humans. This understanding enables the automation of a process that has previously required a human to read and act upon a piece of information. A concrete example would be a cognitive AI in order processing. In order processing, software robotics can export the order information to the ERP system after a human has first read the order and fed the order information to the RPA. In other words, a human is needed to read the information from the order in this process. With advanced, cognitive AI the human can be completely replaced in the process, and the savings are significant compared to using a more common lightweight, non-human-centric AI.
What are the benefits of a truly cognitive AI for business?
At best, using a cognitive AI can lead to cost savings in expert work. High-paid experts won’t need to spend their time on manual work and reading data and are able to use their time more efficiently. Cognitive AI can also enable the better use of existing solutions, rather than replacing them. Making use of knowledge contained in vast amounts of unstructured data may require reading and classifying hundreds or even thousands of documents. This process may not be possible or cost-efficient to implement with human resources. However, through cognitive AI, it is possible and cost-effective task to perform.
How can the competence of an AI provider be verified?
While getting an AI for a project that aims to automate human-generated information, the supplier should be able to answer the following questions.
- Does the AI supplier have experience in automating the process in question?
- Is the supplier’s AI capable of understanding context?
- Does the solution understand human-generated information, in whichever form?
- What level of accuracy will the project help to achieve?
- How quickly and at what cost can a human-centric AI solution be produced?
- Does the supplier offer pre-trained deep models that have already captured vast general and industry knowledge?
- Can the AI make simple decisions or is it able to help experts?
The accuracy of AI should be closer to 100% than 80% or 60%. Solutions should also understand human-generated information, regardless of its format. The starting point for collaboration should always be that the project frees up expert resources to where they are actually needed. The implementation should also be as fast as possible.
For example, traditional text analytics systems can solve almost any problem, provided that you use 1 to 75 man-years to code the algorithms. With a language model-based, deep learning AI you can do this in three days instead of spending years and years developing and then maintaining the model.