Why AI should arbitrate food quality

Quality standards designed for industrial production are eminently unsuitable for natural goods. QC standards as we know them were designed to exclude variability - to ensure that every machine-made product is identical and interchangeable.

While this ensures that all tomatoes that we buy in the supermarkets are near identical; it also ensures that a significant portion of perfectly good tomatoes is rejected on the basis of quality. Supermarkets rightfully assign the blame to consumer preferences.

FoodWise estimates 20% to 40% of fruit and vegetables are rejected before they reach supermarket shelves. This is due stringent regulations on how food looks – a consequence of our unrealistic aesthetic expectations as consumers

Anyone who has eaten a homegrown tomato will know the difference in taste. Those who research this further will learn of the significant differences in nutritional value, but this does not figure in the mass consumer notions of quality - which are primarily aesthetic and intolerant of variations that are natural to the natural world. The problem rests with us, but we add to it through our archaic methods of assessing quality.

Why 18th Century Quality Standards fail the food industry

• Because they are designed from the point of view of the buyers - who are not responsible for food loss along the chain
• They are designed to be used at the purchase point (does not reach into the growing and harvesting cycle)
• They are based on geometry, colour and chemical composition & damage - set within a certain tolerance level
• They are not structured to handle seasonal, regional and genetic variations
• They are not designed for matching consumer demand and supply
• The lack of universal standards due to the high levels of variability and preferences opens up possibilities for disputes and unfair practices

These shortcomings contribute to alarming amounts of food waste globally.

We need better ways of measuring quality

In fact, we had them. We lost them due to the industrialisation of the food industry. If you visit a market fare you will notice the active engagement of buyers in assessing what they buy, by touch, by smell, by colour and price. The QC here is highly aligned with individual choices, preferences and purchasing power. It is highly efficient in matching supply and demand - with minimal waste. The massive efficiency gains that the industrial revolution brought us in terms of productivity came with a price - wastage - estimated to be more than a trillion $.

Can we change this?

The short answer is no - particular in the industrial world. It will take a long time for consumer preferences to change, but we are seeing the beginnings of it. What is possible however is to reduce food loss by using the best computational practices of our times that will enable us to:

• Assess quality spanning starting from growing to consumption
• Positive management of natural variations
• Assess the degradation due to time, storage and transport conditions
• Match consumer preferences to what farmers can grow (instead of interpreted preferences)

We need computational models - not gateways

If we can have a model of a fruit or a vegetable, we can build a better understanding of all its stages of development and all the issues related to those stages. We can know and predict the effect of temperature, humidity and transport conditions. We will know the ideal time for harvest, we will also know which market will pay what price for it. We can most certainly reduce the waste - as the state of the vegetable can now be reliably predicted from harvest to consumption. Growers, transporters and buyer have much to gain if they have a reliable verifiable model of what they are transacting. There is nothing new in this this has happened in almost all industries - except in agriculture.

Where do we start?

Easy things first. Most people in the food industry have a smartphone - which comes with a high-quality camera and connectivity sufficient to do the most complex analysis. The bottleneck so far has been imaging - the inability to obtain reliable geometric, colour and texture information using different phones in different lighting conditions. The technology to trace is now well in place. The two need to come together, along with a compilation of all that we know about food spoilage and its causes - but connected to an open-source computational model. This is something that we do not have yet.

Once we have a model

We can begin to do pretty cool things like:

• Advice farmers what to grow when to maximize their earnings
• Make sales before harvest
• Transactions through electronic negotiations contracts
• Arrange logistics to reduce spoilage during transport
• Accurate instore discounting - sell before it spoils
• Accurate assessment of consumer preferences
• Electronically determined fair pricing based on the state of prodcue

Now about AI?

That is a very big balloon that every industry is beginning to blow. But there are now too many, and that is not a bad thing too, but it can be confusing. AI certainly has the capacity to handle the complexity of the natural world. It has no issues with variations - it thrives on it. It does a great job of assessing consumer preferences. Using it, companies already know more about our tastes and preferences than we know ourselves. AI can certainly help in reducing the colossal waste in the food industry. But AI is fundamentally a computational technology and its latest incarnation it can learn by itself - provided there is sufficient data connecting what's grown to what's consumed harvested along the entire chain. The common approach would be to mine this data mindlessly.

A superior approach would be to create structured open-sourced models of all produce in a way that it contains, geometric, colour, textural, source of origin and nutritional information from its growth stage to consumption - to be a digital twin of the real thing, to eliminate the information loss that it suffers along the way from plant to plate.

Such an approach will increase our understanding of quality all along the chain - so that we may address its deficiencies where we can, instead of controlling quality with colossal waste.

---------------------------------------------------------------------------------------------------------------

Exerts from the ABC SA Country Hour Interview: For full interview :