Extra typically, the AI PPN is sure to be controversial and has already spurred insightful dialogue on LinkedIn. I’d advocate the posts by Kieran McGaughey and Ian Makgill. I provide some further ideas right here and sit up for persevering with the dialog.
For my part, one of many potential points arising from the AI PPN is that it goals to cowl fairly a couple of totally different facets of AI in procurement, in addition to neglecting others. Barely simplifying, there are three broad areas of AI-procurement interplay. First, there may be the difficulty of shopping for AI-based options or companies. Second, there may be the difficulty of tenderers utilizing (generative) AI to put in writing or design their tenders. Third, there may be the difficulty of using AI by contracting authorities, eg in relation to qualitative choice/exclusion, or analysis/award selections. The AI PPN covers facets of . Nonetheless, it isn’t clear to me that these might be handled collectively, as they pose considerably totally different coverage points. I’ll attempt to disentangle them right here.
Shopping for and utilizing AI
Though it primarily cross-refers to the Tips for AI procurement, the AI PPN contains some content material related to the procurement and use of AI when it stresses that ‘Industrial groups ought to pay attention to current steerage when buying AI companies, nevertheless they need to additionally bear in mind that AI and Machine Studying is changing into more and more prevalent within the supply of “non-AI” companies. The place AI is probably going for use within the supply of a service, industrial groups might want to require suppliers to declare this, and supply additional particulars. This can allow industrial groups to contemplate any further due diligence or contractual amendments to handle the impression of AI as a part of the service supply.’ That is an sufficient and probably useful warning. Nonetheless, as mentioned beneath, the PPN suggests a solution to go about it that’s for my part flawed and probably very problematic.
AI-generated tenders
The AI PPN is nevertheless principally involved with using AI for tender technology. It recognises that there ‘are potential advantages to suppliers utilizing AI to develop their bids, enabling them to bid for a larger variety of public contracts. It is very important be aware that suppliers’ use of AI is just not prohibited through the industrial course of however steps needs to be taken to know the dangers related to using AI instruments on this context, as can be the case if a bid author has been utilized by the bidder.’ It signifies some potential steps contracting authorities can take, similar to:
-
‘Asking suppliers to reveal their use of AI within the creation of their tender.’
-
‘Endeavor applicable and proportionate due diligence:
-
If suppliers use AI instruments to create tender responses, further due diligence could also be required to make sure suppliers have the suitable capability and functionality to fulfil the necessities of the contract. Such due diligence needs to be proportionate to any further particular danger posed by way of AI, and will embody web site visits, clarification questions or provider displays.
-
Further due diligence ought to assist to determine the accuracy, robustness and credibility of suppliers’ tenders by using clarifications or requesting further supporting documentation in the identical means contracting authorities would strategy any uncertainty or ambiguity in tenders.’
-
-
‘Doubtlessly permitting extra time within the procurement to permit for due diligence and a rise in volumes of responses.’
-
‘Nearer alignment with inner prospects and supply groups to convey larger experience on the implications and advantages of AI, relative to the subject material of the contract.’
For my part, there are a couple of problematic facets right here. Whereas the AI PPN appears to attempt to not single out using generative AI as probably problematic by equating it to the doable use of (human) bid writers, that is unconvincing. First, as a result of there may be (to my information) no steerage by any means on an evaluation of whether or not bid writers have been used, and since the AI PPN itself doesn’t require disclosure of the engagement of bid writers (o places any thought on the truth that third-party bid writers ma have used AI with out this being identified to the hiring tenderer, which might then require an extension of the disclosure of AI use additional down the tender technology chain). Second, as a result of the strategy taken within the AI PP appears to level at potential issues with using (exterior, third-party) bid writers, whereas it doesn’t appear to object to using (in-house) bid writers, probably by a lot bigger financial operators, which appears to presumptively not generate points. Third, and most significantly, as a result of it reveals that maybe not sufficient has been finished to this point to sort out the potential deceit or provision of deceptive info in tenders if contracting authorities should now begin fascinated with methods to get expert-based evaluation of tenders, or develop fact-checking mechanisms to make sure bids are truthful. You’ll have thought that whatever the origin of a young, contracting authorities ought to be capable to test their content material to an sufficient stage of due diligence already.
In any case, the most important challenge with the AI PPN is the way it suggests contracting authorities ought to take care of this challenge, as mentioned beneath.
AI-based assessments
The AI PPN additionally means that contracting authorities needs to be ‘Planning for a basic enhance in exercise as suppliers might use AI to streamline or automate their processes and enhance their bid writing functionality and capability resulting in a rise in clarification questions and tender responses.’ One of many potentialities might be for contracting authorities to ‘struggle fireplace with fireplace’ and likewise deploy generative AI (eg to make summaries, to scan for errors, and many others). Curiously, although, the AI PPN doesn’t straight check with the potential use of (generative) AI by contracting authorities.
Whereas it features a reference in Annex A to the Generative AI framework for HM Authorities, that doc doesn’t particularly tackle using generative AI to handle procurement processes (and what it says about shopping for generative AI is redundant given the opposite steerage within the Annex). For my part, the generative AI framework pushes strongly in opposition to using AI in procurement when it identifies a sequence of use instances to keep away from (web page 18) that embody contexts the place high-accuracy and high-explainability are required. If that is the federal government’s (justified) view, then the AI PPN has been a missed alternative to say this extra clearly and straight.
The broader challenge of confidential, categorized or proprietary info
Each in relation to the procurement and use of AI, and using AI for tender technology, the AI PPN stresses that it might be essential:
-
‘Setting up proportionate controls to make sure bidders don’t use confidential contracting authority info, or info not already within the public area as coaching information for AI programs e.g. utilizing confidential Authorities tender paperwork to coach AI or Giant Language Fashions to create future tender responses.‘; and that
-
‘In sure procurements the place there are nationwide safety issues in relation to make use of of AI by suppliers, there could also be further issues and danger mitigations which are required. In such situations, industrial groups ought to interact with their Data Assurance and Safety colleagues, earlier than launching the procurement, to make sure proportionate danger mitigations are applied.’
These are points that may simply exceed the technical capabilities of most contracting authorities. It is extremely arduous to know what information has been used to coach a mannequin and financial operators utilizing ‘off-the-shelf’ generative AI options will hardly be able to evaluate themselves, or present any significant info, to contracting authorities. Whereas there might be contractual constraints on using info and information generated underneath a given contract, it’s rather more difficult to evaluate whether or not info and information has been inappropriately used at a unique hyperlink of more and more advanced digital provide chains. And, in any case, this isn’t solely a difficulty for future contracts. Information and data generated underneath contracts already in place will not be topic to sufficient information governance frameworks. It could appear {that a} extra muscular strategy to auditing information governance points could also be required, and that this shouldn’t be devolved to the procurement perform.
Find out how to take care of it? — or the place the PPN goes flawed
The largest weak point within the AI PPN is in the way it suggests contracting authorities ought to take care of the difficulty of generative AI. For my part, it will get it flawed in two alternative ways. First, by asking for an excessive amount of non-scored info the place contracting authorities are unlikely to have the ability to act on it with out breaching procurement and good administration rules. Second, by asking for too little non-scored info that contracting authorities are underneath an obligation to attain.
An excessive amount of info
The AI PPN contains two potential (different) disclosure questions in relation to using generative AI in tender writing (see beneath Q1 and Q2).