LOOK INSIDE THE COLLAR of a Patagonia Responsibili-Tee t-shirt and you’ll find a set of metrics. This t-shirt is made of 4.8 plastic bottles plus 0.26 pounds of cotton scraps, it reads. In addition, you’ll find it uses 96% less water and produces 45% less carbon dioxide compared to a conventional t-shirt. 

“Is each product worth the environmental cost?” the California outdoor clothing company asks in its annual report. For Patagonia, even the shirt on your back is a multi-objective problem: How can it reduce the carbon, water and waste costs for every product?

“While we do have specific milestones and goals in place, we’re also working on comprehensive changes that we hope can influence the industry,” Patagonia’s Environmental Activism Manager, Meghan Wolf, told the Vegas Business Digest. “We’ve instituted a metric called the Environmental Profit & Loss (EP&L), which calculates carbon, water and waste costs of every item we sell, and it’s something we reference when making decisions on improvements – or reductions – on what we make.”

Ethical consumers have long raised concerns about the humanitarian and environmental impact of clothing production but in recent years an increased scrutiny of brands’ supply chains has become more widespread. It’s also a curious challenge for artificial intelligence, which excels at handling multiple conflicting goals by efficiently evaluating vast data, and using clever optimisation techniques without assumptions.

Awareness is growing amongst Gen Z and millennials about environmental and social issues, including supply chain emissions – or “Scope 3” emissions which companies are indirectly responsible for and can account for 85% of a brand’s carbon footprint – and increased brandresponsibility, says Tamara Stark, Campaigns Director at Canopy, a non-profit which works with companies including Ben & Jerry’s, Lush, H&M, J.Crew, and Zara to achieve deforestation-free supply chains.

“Might all clothing labels of the future look like Patagonia’s? And since AI is constantly improving our ability to manage multi-objective problems, can neural networks help us get there?”

“These mindset and behavioural shifts are propelled by a heightened awareness of the urgency of both the climate crisis and the spiralling impact on species and biodiversity,” Stark says. “Consumers, brands, and non-profits are all playing a role in advocating for more sustainable and responsible supply chains, and their collective efforts are contributing to the growing focus on transparency, accountability, and sustainability in global supply chains.”

Recognising we need supply chain transparency is one thing but creating a reliable and accountable system to achieve that is another. How should it be measured and standardised across companies in the global marketplace? How would that information be clearly communicated to consumers? And how could brands use that information to further reduce their impact? 

For Stark, there are a number of criteria beyond just measuring carbon emissions that can be useful for consumers, and brands themselves, when quantifying supply chains and assessing their environmental impact.

“Metrics such as ensuring there is no ancient and endangered forest fibre in the product as a minimum for market entry is key,” she says. “Other indicators could include the number of hectares of forests protected, the reduction in deforestation and degradation rates compared to industry benchmarks, species viability and richness, biodiversity index, water footprint and right-sizing of packaging.”

Over 150,000 B Corp businesses voluntarily take part in B Lab’s Impact Assessment which provides a measurable and detailed environmental and social impact score, across five categories including governance, workers, and the environment, albeit one you have to search up online.

Might all clothing labels of the future look like Patagonia’s? And since AI is constantly improving our ability to manage multi-objective problems, can neural networks help us get there?

“We’re at a point now where we have a level of computing power that enables us to do calculations on a scale that was unimaginable before,” says Rebecca Jeffers, a senior AI Research Engineer working on DeepPack, which aims to reduce carbon in the supply chain through more efficient packing and transport.

“Supply chains are complex environments where you have lots of data and lots of decisions to make, and any decision you make today will affect the decisions you’re able to make tomorrow,” she says. “Using AI allows us to make use of all the data that is available from all these different parts of the supply chain.”

And when that computational power is combined with intelligent algorithms we can do even more, says Jeffers. “We’re no longer talking about coding a set of laws that your computer then follows, we actually want the computer to learn itself as well.”

The way this might work in practice, according to Jeffers, is once a brand or industry body has chosen a series of environmental and social metrics, such as carbon emissions, waste and water usage, you would then shape the agent’s reward signals in a specific way.

“If one of the indicators was more important than the others, you would just put a heavy weight on that indicator,” she says. “So, for example you might decide it is ten times more important to have a low deforestation metric than it is to reduce the transport emissions, so you’d multiply the deforestation metric by 10.”

“The agent is learning based on the reward signal, so the real challenge is making sure you are defining your reward signal in a way that results in behaviour that is aligned with what you actually want. Your neural network is giving you a policy to follow in effect.”

Could you be strict with your agent and programme it to not go below a certain score for all your metrics, to achieve a low environmental and social impact score across the board? You could try, Jeffers says, but you’d also be restricting your agent’s ability to learn novel strategies that you as a human might not have thought of. And it also could create an unsolvable problem, in which case you might have to change the constraints you’re imposing on the agent.

“You need to be able to correctly measure the consequences. Are the metrics you’re following just vanity metrics? Or do they actually mean something?”

When it comes to using AI modelling to make supply chains more sustainable the biggest challenge for Jeffers boils down to the integrity of the data. “You need to be able to feed your AI reliable, accurate and useful data,” she says. “A saying that is often used is: ‘Garbage in, garbage out,’ so if you’re feeding it data that is faulty, what is going to come out will be faulty.”

Stark believes rating systems, such as scores out of 10, can be used to assess a supply chain’s performance on each criterion and indicator and provide a quantitative measure of its nature-positive impact. But she says: “It’s important to develop robust and scientifically-sound methodologies for data collection, analysis, and verification to ensure the accuracy and reliability of the assessment.”

Either you will have a lot of real-world data where actions were taken, and consequences recorded, so you’re able to learn from that, Jeffers says, or you train your agent in a simulated environment, but if that’s the case, you need to make sure your simulator is representing the real world. And with problems that are a bit more complicated, you may not be able to generate fake data in the same way.

She warns: “You need to be able to correctly measure the consequences. Are the metrics you’re following just vanity metrics? Or do they actually mean something?” It’s still all very well knowing how to code and doing maths and statistics but unless you actually understand the problems you’re trying to solve, it’s kind of useless,” says Jeffers.

Jeffers notes that even when giving AI certain objectives in terms of metrics it won’t always result in the type of behaviour you want to see. “We could give incentives around carbon footprint for example and maybe that would incentivise companies to really invest in carbon offsetting without actually making any internal or structural change,” she says. “Or around land use being used for forestation which could perhaps put local farmers out of work.”

It’s a reminder, she says, not to expect too much from the AI and to remember it’s only as good as the model we’re asking it to design. It’s also another reason to make sure you have a diverse workforce when you’re working in AI. “We all have our blind spots which we can hopefully minimise when working as a collective,”Jeffers says.

Yossi Sheffi, Director of the MIT Centre for Transportation and Logistics, and author of The Magic Conveyor Belt: Supply Chains, AI and the Future of Work is sceptical as to whether brands would want to use AI to make their supply chains more sustainable and transparent. “Is big business serious about it?” he says. “And would consumers be willing to pay for it?” 

He says research shows only 7% of consumers will choose sustainable products over regular products and that businesses cannot start investing in tech and changing their supply chains if consumers are not willing to play their part.

Sheffi cites the example of the supermarket Tesco, who as far back as 2008 trialled a carbon footprint labelling system on their own brand products, but then dropped the scheme in 2012 claiming it was too expensive and time-consuming to calculate each product’s rating. Customers were also confused as to whether a high or low rating was desirable. “It bombed,” Sheffi says.

Even if you could get the brands on board with a globally recognised certification system, connecting an inordinate amount of supply chain data (which Sheffi believes has “exactly zero” chance of happening, in part because many of the brands would consider this data proprietary), he is not convinced the data would be trustworthy through all the tiers of a supply chain.

“The AI can’t go out and make this data magically,” he says, “humans have to accurately report

what’s happening. Sheffi was working with a technology company that was trying to get rid of conflict minerals from its supply chain, but the mines were at tier 12 of their supply chain, with many intermediary individuals and companies playing a role in between. Eventually they did find a solution, but it took four years and required getting 400 other companies on board, which illustrates the complexity and interconnected nature of the global supply chain network.

“Businesses cannot start investing in tech and changing their supply chains if consumers are not willing to play their part.”

Stark is less cynical about the future of transparent supply chains. She believes there is strong market demand for ethical supply chains and that brands will partner with non-profits to achieve it, not least because it’s what consumers want. Peter Grimvall, who heads up Ikea’s

supply chain design and planning, agrees: “Our customers will lead the way. And if they say: ‘We would like to know this’ or ‘We would like to have that,’ then for sure we will go in that Direction.”

Google’s Global Director of Strategic Business Transformation Louisa Loran sees collaboration and the sharing of data – and its role in making the supply chain run more efficiently – as something that will reshape how we make, move and buy things in the future. Sharing data and the learning and productivity gains it brings will be essential to remove waste and make industry more sustainable, she says.

“If the environment becomes a bigger part of decisions than what it is right now, there is a potential to force change. Most changes are anchored in money, regulation or masses of consumers,” Loran says. “Consumers must be mobilised to help drive change.”

Take a bottle of water for instance, she says, if she could compare the CO2 cost of one brand versus another, she could not only make a more informed decision, but get a better picture of her own aggregate carbon footprint and take responsibility for her choices elsewhere as well.

“But today I am not informed enough to know the impact and consequences of my (personal) supply chain as it doesn’t really exist yet – but it is very possible with data and AI,” Loran says. “Someone just has to make it easier for people to engage with and then we will be driving change together.”