Artificial additives: Artificial Intelligence in Agriculture

The world is increasingly a digital space. Between 1995 and 2017, the number of people using the internet has grown from 16 million to 3.7 billion, of which 2 billion are from countries traditionally defined as ‘developing.’ The possibilities for creating and sharing data is incredible, and has real potential for improving project impacts and adoption, via i) larger audiences for agricultural extension, ii) Lower-cost outreach, iii) Accelerated data collection and interpretation.

The creation of all this data however still requires interpretation, which usually requires pain-staking hours of work.  The problem to date with using machines to automate previously human tasks is that machines are fundamentally dumb, being programmed to do one task, seeking specific pre-determined inputs and limited by the capability and foresight of the designer(s). However, the growing rise of AI may change this.

AI, is not only a three toed sloth, but refers to ‘artificial intelligence’ that utilizes ‘adaptive learning systems’ to new knowledge, interacting with and interpreting information. With each interaction the AI gets smarter, predicting responses, testing and then identifying changes to parameters to find the desired response. These traits in a human are often referred to as “deductive reasoning, planning, learning, perception” (etc…), so we really need to start being nice to our computers. Simply put, machines that learn from a range of interactions it has received and memorizing what caused different responses to adapt for the desired result. Sounds pretty smart!

Figure 1 Ai, coming to disrupt the way we run projects… source: birdquest

A quick search of novel solutions to on-going problems in agricultural development will find a search field awash with innovations applying aspects of AI:

• automated extension: advice tailored to your crops, location, resources using smart and adaptive step logic
• rapid disease diagnosis: use of phones to take photos of suspicious crop or livestock changes to then be automatically assessed for known diseases and/or identify new ones
• Affordable mapping: attaching increasingly cheap and powerful smartphones onto helium balloons attached with a string to take photos of land to determine soil condition

Development organizations such as the Bill and Melinda Gates Foundation and USAID are investing heavily in innovative, technology that apply aspects of AI (Socialcops – Big data, Crop Manager – extension, Digitalgreen – community led extension, iCow – extension and record keeping, radiant.earth – low-cost imagery, KUMU – stakeholder mapping). There is a good argument for agricultural research for development to invest greater skills and money in this field if we are to adapt to the future. And of course, applying AI systems effectively requires more than grafting western designed systems to in-country context – but working with partners to identify genuine problems to add value. So we don’t need to get worried about our jobs just yet, and if you are, here’s an article about Universal basic income to get you excited (or just learn to code).

Steps for the future:

1. Seeking, mapping and brokering partnerships with innovative IT upstarts with coding skills and in-country partners
2. Hiring research experts in coding and AI specific technologies
3. What else?

It is important to note that it is more than just plonking AI in a country, you need to train people, make sure it’s a value add and that the stakeholders see it as a value add. Then it needs to be resourced beyond the life of the project.

Aid projects are generally responsive to partner countries priorities, and as of yet I haven’t heard any clamoring for help with implementing AI systems in agricultural contexts. However,

While ACIAR doesn’t work in intelligent computer systems, we certainly do have a number of non-specialist jobs that are begging to be automated (provided the task can be completed as well as a human) (and projects that could benefit from improved access to information and extension – freed from the restriction of underfunded partners).

Examples of areas begging for smart automation:

• Writing of contracts and legal agreements
• Personalized extension – often massively underfunded in developing nations
• Trawling global big data to identify best practice extension methods
• Improved accuracy of weather predictions and associated likelihood of crop success
• Recognition technology utilized by Google Lens: plant and livestock deficiencies and diseases

 

Views expressed in this blog are the authors own.