How do you ensure that Artificial Intelligence (AI) results are free of errors or with minimal errors?
Use AI decision-making tools with defined algorithms that automate micro-decisions, particularly for real-time and higher-volume decisions.26%
Ensure model reliability, trustworthiness, security, and privacy.50%
Combination of technical and managerial approaches10%
All of the above14%
Sort by:
There is no way to ensure AI is free of errors, but from a minimal error standpoint I think its important to consider things like cross-validation, testing on diverse datasets, and comprehensive error analysis. In the healthcare world I am in we focus a lot on model bias as well. Healthcare data is often really messy so if AI is being trained without regard to data curation and validation it is likely going to be extremely biased.
I agree with you and one of my discussion was:<br>How has the pharmaceutical industry been affected by the integration of artificial intelligence (AI) into research and development (R&D)? | Gartner Peer Community<br><br>
You don’t. You accept that these things are imperfect and non-deterministic. You can reduce the error rate by doing the same work on multiple models and looking for agreement / disagreement, just as modern passenger jets do with three dissimilar flight computers. “Free of errors” is like saying a car “free of crashes” - these things will never be true but we can and should aim to get as close to perfection as possible over time without stopping progress along the way.
It is challenging to produce 100% error-free results using AI-generated content in general. Even with the most modern technologies and processes, there is still the possibility of human error or other unforeseen events leading to errors. However, AI-generated material can be extremely beneficial in a variety of domains, including natural language processing, image identification, and predictive analytics.
How do you ensure Human Intelligence (HI) results are free of errors or with minimal errors? Make sure your AI or HI is educated with high fidelity inputs, require backup (documentation, supporting evidence), check with more than one source for consistency, estimate likelihood of accuracy, and determine whether to use the information based on projected benefit and risk.