What is Farming Co-intelligence?
Peter Gill
Managing Director
What is Co-intelligence?
Generative AI is the most remarkable human-created phenomenon of all time. Period.
"Why sometimes I've believed six impossible things before breakfast," says the White Queen in Lewis Carroll's Alice Through the Looking-Glass. Yet now through AI we have the option of being able to experience a million and one previously impossible things before breakfast: interactive tutoring in Mandarin Chinese, diagnoses of medical symptoms superior on average to those of the average doc, and instant access to impressive knowledge of deeply technical areas of the whole of world agriculture.
Understanding Generative AI is a growing science; even its creators wouldn't be able to create a definitive manual. Understanding AI as a subject is happening mostly through expert users' experiences and their sharing of new knowledge.
An interesting observation is the existence of what may best be described as AI personalities (each platform having its own). An oft-observed consequence of this is that when working with Generative AI it is kind of just human nature to slip into believing that you are working with a living, breathing person. It's harder to step away and do the reality check of remembering one is chatting, or debating, with a supercluster of Nvidia H100 8-processor clusters residing in an air-conditioned facility on a different continent.
The different AI personalities seem friendly enough, the movie characters HAL, Marvin the Paranoid Android and Jarvis may come to mind, yet no-one scripted or programmed the Generative AI personalities - how they have come about 'naturally', and may evolve in the future, is one aspect of future Generative AI study.
The Great AI Paradox
There is one essential thing that every AI users needs to have permanently front of mind. I think of it as the Great AI Paradox. In a nutshell: in a few seconds Generative AI can recall valuable minutiae from its training run and perform incredible writing or calculation or artwork tasks and generally dazzle us humans with its ability to succeed in tasks we find time-consuming and/or difficult. Yet what the friendly AI personality does not possess is any idea - not the foggiest of clues - of how great it is doing. To quote Dr Andrej Karpathy, an AI guru's guru, with my added underscore: "GPTs don't want to succeed". AI is as conscious as your teapot.
The fact that they do succeed so highly remarkably and so consistently changes nothing. Successful co-intelligence users won't be lulled by stunning and consistent success and always work on with the AI acting as its critical friend. Whatever a particular output may have been the onus stays with the user to get you both where you need to be. Thinking of the football manager's proverbial request for her players to give 110% so it is with AI: always squeezing for more is the co-intelligent user's best strategy.
.
The diagram illustrates the co-intelligent relationship between users and AI systems. To the left are your instructions. On the right are three possible levels of initial output quality i.e. correctness in relation to what you want to do.
Increasing quality - when you haven't immediately hit on what you consider to be excellent - depends on two things.
Firstly knowing which quality level the initial output is in. Only you can ever know, as having an opinion on this is above the AI's paygrade. The level of a user's subject knowledge can make an enormous difference to AI performance: getting the most from co-intelligence is teamwork. Secondly an understanding as to why the AI's output wasn't good enough, or was absolutely brilliant. This requires some understanding of AI 'thinking' (which is not comparable to our thinking) plus good knowledge of the techniques - of which there are many - that may tune the AI's 'thinking' to change a particular output quality.
So, here are four tenets of Co-Intelligent AI:
-
Co-intelligence is a never-ending journey not a destination
-
AI doesn't know what is 'right' and what is 'wrong'
-
Users knowledgeable in a subject will, in general, always achieve the highest quality output from AI
-
Depth of knowledge of AI principles and techniques is essential equally to subject-experts and non-experts.
Farming Co-intelligence
Businesses of all kinds both within and outside of agriculture can be beneficiaries of the transformative technology that is Generative AI.
In AI jargon the term 'use cases' means the things AI is used for but for the many new users the non-existence of a user manual is problematic. In Rumsfeld Matrix terms, the great majority of AI's functionality for a new user comprises unknown unknowns. Weird as it may seem to us even the creators of Generative AI aren't aware of all the functional capabilities they have set in motion: some are in use and some await discovery.
To fill the knowledge void for users in and around agriculture our AI training is designed to have new users on the right tracks with regard to farming co-intelligence use cases within agricultural AI, farm management AI, business management AI, agronomy AI, animal health AI, professional services AI and marketing AI as well, of course, as covering the general principles needed to understand Generative AI and the AI techniques needed to produce optimal outputs.
A newer functionality is being able to share AI-created 'apps' which phenomenally even further extends the power and value of Generative AI.
As an example, with AI sharing, a farmer-user or farmer-supplier who has developed a neat AI app in her or his farm office can give it (or sell it) to anyone else. The value of AI app transferability is easy to see:- (a) no re-inventions of the wheel, (b) new users able to get immediate value from AI, and (c) the potential for co-developments extending the functionalities of the original app. Watch this space I'd say.