The breaking national news today was extraordinary. The West Midlands chief of police finally admitted that he used Artificial Intelligence to create an incorrect report which was used to make a critical decision and which will inevitably now lead to his resignation. AI can be a great tool, however, at Sustainable Solutions, we have severe reservations and therefore have created a company policy on its use. In summary we can categorically claim and reassure our clients that we DO NOT use AI in any way for the generation of our high-level reports such as DSEAR, CDOIF, Fire Risk assessments, COSHH risk assessments etc. these are some of the reasons:
- The fear of embedding the mediocrity produced by AI and in some cases the hallucinogenic effect which has been noticed by its use at a personal level. We tested the technology on a sample NEBOSH exam question and one out of the nine answers it provided was categorically wrong. The realisation that it could be easy for some unscrupulous person to intentionally feed the technology with incorrect information and pollute all possible future answers.
- The concern that our high-end technical reports become a learning tool which will then help to refine all of our competitors trying to write similar reports.
- In some cases, the concern that we are breaching both confidentiality protocols and possibly GDPR parameters. (We are on the radar of the Information Commissioners Office – ICO)
- The omission of the opportunity for the skilled and experienced assessor to have a much more nuanced overview or recognition of subjective pinch points which just would not be picked up by AI unless an extreme level of input detail was provided.
AI software is an incredibly powerful tool, and we do allow the use of it in certain situations, usually to scope out and gain an insight of the possible problems we are liable to encounter before we go to a site. Individual paragraphs are subject to spell and grammar checks, if the passage does not contain anything sensitive. But on the whole, and for the moment we continue to invest the effort to complete proper and thorough research or apply basic scientific principles when developing our reports.
For example, we are next due to go to site to complete a DSEAR risk assessment for a company using acetylene gas, piped into a small workshop where they are shaping glass ampules. Apart from the application of various technical references on the HSE website, technical references provided by the compressed gas supplier, we will also be referencing British Compressed Gas Association CP5 and CP6 standards as well as completing zone calculations based on IEC 60079-1. Yes we could use AI to run those hazard zones, but the risk of being deceived and getting it wrong could be catastrophic. The confidence levels are just not high enough.
