1
Connecting the next wave of IoT Devices
2
Changing the face of the IoT ecosystem
3
Enterprises turn to IoT for sustainability and profits
4
AI at the IoT edge
5
Operationalizing AI
6
AI Vanishes to go Mainstream
7
AI Growing Up and Learning Accountability
8
Merging Tech
9
Quantum computing moving beyond the hype
READ MORE
Given the enormous strides made in AI applications in the last few years and given its substantial energy consumption and environmental impact, it’s now time to ask how we can sustain AI’s growth without demolishing resource sustainability.
AI SUSTAINABILITY: TWO SIDES OF A COIN
Applied Intelligence Group
As we reach the halfway point of 2022, we’re now living in a world where global developments mean we rely on the power of transformative technologies such as AI, IoT, and Quantum, to provide solutions more than ever. The adoption and continued progression of these technologies are essential across a wide range of industries and geographies, that need support to overcome challenges faced. The implementation and acceleration of transformative technology means projects can move forward at pace, and this is where we will really start to see a change in the current status quo. Individually both AI and IoT are valuable, but the power of these technologies when they come together is where we will continue to see the biggest impact that will make a real difference. Their confluence will turn what were simple solutions into complex, truly impactful offerings that will alter society and businesses for years to come. Individually both AI and IoT are valuable, but the power of these technologies when they come together is where we will continue to see the biggest impact that will make a real difference. Their confluence will turn what were simple solutions into complex, truly impactful offerings that will alter society and businesses for years to come. It's the continued advancements of each of these technologies that will enhance the other. Omdia, our technology research brand, estimates that by 2030 there will be 75 billion IoT devices around the globe, which will provide an unimaginable amount of data to strengthen AI systems. At the same time, AI allows this vast amount of data to be analyzed and acted upon with unprecedented speed. As we move forward, organizations will be looking at how they can gain competitive advantages through leveraging quantum computing to move beyond the limitations that traditional architecture presents; and increasing investment in AI-driven automation. Alongside this, also ensuring they converge Subject Matter Experts with Data Scientists and Engineers throughout various levels of the process. Whilst there is significant momentum in the space, Omdia predicts there are areas that still need to be addressed across 2022 and beyond. These include interoperability with IT existing systems, industry-wide standardization of AI measurement and success; implementing standardized regulation and governance; and a greater drive to improve diversity within the sector.
Jenalea Howell Vice President, Applied Intelligence Group at Informa Tech
Building a winning AI strategy doesn't have to break the bank, or the planet.
Brought to you by
Introduction: conflicting objectives Achieving both objectives AI applications for sustainability How environmental impact is measured Conclusion
Contents
Next
In the last five years, the field of AI has made huge strides in speech recognition, natural language processing, decision-making, and the integration of image recognition with data search and correlation. These advances have powered new breakthrough applications, including language translation, autonomous driving, security systems, and generative transformations - many arising from pre-trained, foundational Machine Learning (ML) models found within generative AI solutions such as OpenAI’s ChatGPT. Not surprisingly, market penetration for AI has shifted from the 13% early adopter phase to the 34% early majority. At the same time, however, these new applications consume many times more compute cycles than “traditional IT” compute consumers. This raises three major questions:
Introduction
For example, AI’s predictive capabilities as used in intelligent energy grids promise to more efficiently manage the supply and demand of renewable energy. Similarly, AI is used to analyze real-time satellite imagery to measure use changes, crop failures, and the impact of natural disasters. AI-augmented agriculture then employs this data to guide planting patterns and program agricultural robotics. Undoubtedly AI will find additional new applications that provide visible benefits to society. Corporate AI isn’t even the largest energy consumer. In support of data sovereignty requirements, cloud computing deployments often dedicate thousands of chips inside thousands of server racks that are centralized in regionally isolated data centers around the planet to train AI models, creating two distinct problems, one concerning environmental sustainability and another dealing with AI sustainability itself. Regardless, all AI computation consumes energy and – nearly as important – converts virtually all of that energy to heat, which must be dissipated in the environment.
HOME
Can AI help mitigate that impact by solving related problems, such as energy sustainability, carbon emissions, inefficient logistics, and production waste generation?
Do environmental burdens created by massive AI computation outweigh the direct benefits of AI applications to improve overall productivity?
What is accelerating AI computation’s impact on the environment?
On the other hand, the computation required to train a single average-sized ML model expends enormous amounts of energy: more electricity than 100 US homes use in an entire year. For instance, training OpenAI’s Chat-GPT3 alone used 1.287 gigawatt-hours of energy, or the equivalent of 1000 300W NVIDEO Tesla V100 GPUs over slightly more than a month.
Artificial Intelligence Is Booming— So Is Its Carbon Footprint.
In fact, this same Omdia survey found that the top three reasons for embracing sustainability initiatives are to reduce company costs, meet customer expectations, and aid company growth. Achieving a measurable environmental impact, and actual energy and water conservation, were fourth and fifth place objectives.
Achieving Both Objectives
According to Harvard Business Review, an organization’s sustainability posture requires expertise that leads to the creation of a new C-level position, the Chief Sustainability Officer (CSO). On the other hand, in its 2022 survey of senior data and technology executives, NewVantage Partners reports that 92% of large companies say they are achieving returns on their data and AI investments. That’s up markedly from 48% in 2017. Less than a quarter of companies surveyed in 2023 by Informa Tech's leading research team at Omdia, have a well-defined strategy for sustainability that is tied to the company strategy in terms of actual IT investment.
AI for sustainability concerns exploiting AI to achieve sustainable development goals in water, air, and energy conservation. While companies see AI as a potential sustainability enabler, not every organization’s mission is concerned with sustainability. Yet to be sustainable in a truly impactful manner requires direct financial ties to department or corporate performance measures. Few companies have talent onboard able to do this, and unfortunately many are under the impression that using greenwashing techniques – such as purchasing CO2 offsets – meets this objective.
Many companies already employ AI as part of their business intelligence stack, giving them a leg up in harnessing AI for sustainability goals in terms of possessing the requisite skill and experience in deploying AI regardless of use case. An obvious question is whether companies incorporate environmental costs into their year-over-year AI ROI, which is necessary to measure the true investment before calculating a company-wide return ratio. One of the largest driving ROI factors for AI is finding a balance between time to deployment and long-term cost–spending perhaps more on expensive, high-heat GPUs to speed model training, but falling back to cheaper, low-heat CPUs for inference over the long haul.
Gender gaps on boards
Personal characteristics and promotion
Career progression for women
Job gains
Gender gap/quotes in publications – and subsequent bias in solutions
Contact us
SCROLL
By 2025, without sustainable AI practices, AI will consume more energy than the human workforce, significantly offsetting carbon-zero gains.
REASON 1
REASON 2
REASON 3
reduce company costs
meet customer expectations
aid company growth
Given current energy use trends in AI, the rush to explore new use cases in GenAI while also increasing profits through all forms of AI could well cause a new environmental crisis, as tens of thousands of new, competitive, ML training and inferencing projects get underway in the coming years. Gartner predicts:
“
of large companies say they are achieving returns on their data and AI investments
92%
Companies must think about and approach AI for sustainability and sustainable AI in very different ways.
Back
The 8 Responsibilities of Chief Sustainability Officers
We have to ask about emotional AI and crime, Springer Verlag Open Access, Published: 05 May 2022
Reducing energy consumption. Worldwide, renewable energy accounts for less than 15% of the world’s electrical generation. Thus reducing energy consumption directly contributes to the sustainability footprint of an organization. A 2021 University of Taipei, Taiwan research paper discusses a universal AI workflow aimed at identifying power control tools that reduce energy usage. Based on a survey of data from various studies, the developed workflow identifies energy-saving analytics for applications germane to a wide range of industries. Improving air quality. By analyzing data from air quality monitors and then correlating that data with related energy consumption, AI can assess the impact of air quality on the environment and guide protective policy development. AI-augmented HVAC systems can dynamically adjust heating, cooling, and outside air distribution using real-time air quality data, to optimize energy consumption while lowering operating costs. For example, a company can employ AI predictive models to forecast trends in pollutant spread and concentration, helping focus policy enforcement while ensuring employee safety. The industry journal Applied Thermal Engineering published a meta study in 2023, describing how to design AI functions for energy-efficient HVAC systems, based on reviews of famous companies – such as Google, IBM, Johnson Controls, and Samsung – reporting quantitative energy-saving effects from AI. Optimizing logistics. The distribution and logistics operations of many businesses account the major part of the total corporate carbon footprint. The non-profit Carbon Disclosure Project, in its 2020 Global Supply Chain Report, states that organizations' supply chains often account for more than 90% of their carbon emissions, when taking into account their overall climate impacts. Thus this is an area ripe for AI optimization. Supply chain analytics help ML models forecast required stock levels based on identified demand cycles and early warning of demand changes. AI-driven warehouse management has proven effective at reducing an organization’s carbon footprint through the use of robotic parts carriers and predictive maintenance for those carriers. ML Models can calculate optimized routes for product delivery, taking into account real-time weather and traffic data, with sustainability as a key objective. Eliminating defective production waste streams. Product waste occurs all along the supply chain, from parts to finished goods, including customer returns. According to Forbes, the common practice of free returns has a hidden carbon cost most companies overlook.
AI Applications for Sustainability
50%
Governments, public utilities, manufacturing, and distribution, are already applying AI to sustainability problems in at least six major areas.
equality
1.
2.
3.
4.
5.
6.
6 Transparency to Transformation: A Chain Reaction. CDP Global Supply Chain Report 2020 https://cdn.cdp.net/cdp-production/cms/reports/documents/000/005/554/original/CDP_SC_Report_2020.pdf?1614160765
Couriers typically use heavy polluting vans to come and collect returns. A 10% reduction in returns would save enough energy to power about 57,000 US homes for one year.
“Couriers typically use heavy polluting vans to come and collect returns,” says Forbes. “Factoring in the amount that a van can carry at once, and the fact that some vans are more efficient than the larger transits, a single item using a courier emits 181g of CO2, when being returned.” The article goes on to note that a 10% reduction in returns would save enough energy to power about 57,000 US homes for one year. Many returns could be avoided with better quality control (QC), and AI can play a useful role here at the point of manufacture. ML-based computer vision QC systems on a packaging line can detect flawed products before shipment, letting the order be reworked at its source. The cost of such systems might well be borne by the savings in reducing expensive product return truck rolls.
Universal workflow of artificial intelligence for energy saving, 2021,
Artificial Intelligence Enabled Energy-Efficient Heating, Ventilation and Air Conditioning System: Design, Analysis and Necessary Hardware Upgrades, 2023
Using artificial intelligence to create a tsunami early warning system, 25 April 2023, Cardiff University News
8 How AI can enable a Sustainable Future, 2019,
7 There Is No Such Thing As A Free Return, 2019
Scope 1 covers direction emissions from company owned machinery, factories, and vehicles.
Scope 2 covers indirect emissions sources associated with the generation of electricity, heat, steam and/or cooling.
Scope 3 covers business travel, commuting, waste, and third party deliveries.
Identifying scope 3 risks is more complex than identifying scope 1 and 2. Large language models (LLMs), such as Open AI GPT, can help companies better understand these risks by analyzing vast amounts of online text, such as social media posts, industry reports, research papers, and news to surface insights that might otherwise remain hidden from view.
Predicting, preparing for, and recovering from natural disasters. Natural disasters affecting any given company are thankfully rare, but it’s an unhappy truth that natural disasters happen every day somewhere. Climate change already has altered weather patterns and flood zones, creating new extreme events that impact company operations. Disaster-impacted companies must reprioritize tasks just to stay in business, which means sustainability measures often go out the window. UK professional services firm PwC, in conjunction with Microsoft, published a 2019 whitepaper documenting its research on AI contributions to sustainability. For example, says the report, “An estimated 250 million people are already affected by flooding annually. Our analysis estimates that AI-enabled improvements to forecasting could enable flood early warning systems which would save over 3,000 lives, result in 1.2m fewer people made homeless and mitigate $14m economic damages between now and 2030.” While weather forecasting itself uses basic mathematical methods rather than AI, businesses can exploit AI-based early warning systems to help mitigate the effects of a disaster and return to normal operations. For example, researchers at Cardiff University in the UK and the University of California, Los Angeles and have developed an early warning system combining state-of-art acoustic technology and ML models to immediately classify earthquakes and assess potential tsunami risks. Earthquakes cannot as yet be predicted, but tsunamis occur minutes or hours later, and are amenable to ML prediction. Analyzing supply chain greenhouse gas generation. Compliance standards identify three scopes of risk for greenhouse gas generation (GHG):
Consider How Environmental Impact is Measured
Most businesses know the labor costs of any IT project, but few track the energy cost. Energy costs are based on IT resource energy consumption, which translates directly to carbon footprint and BTU output. Companies will likely need to add monitoring to their AI build processes to capture energy consumption data. For example, in its article How to Make Generative AI Greener, Harvard Business Review authors Ajay Kumar and Tom Davenport note three distinct carbon footprint metrics to consider:
Models having more parameters and larger training data sets consume more energy during both training and inferencing, and when that energy is non-renewable, it generates more carbon. Monitoring data should be brought into a Business Intelligence (BI) dashboard where it’s visible across the organization. Cloud AI providers should provide a similar dashboard for cloud-resident AI training data and workloads. Organizations can use validated Machine Learning CO2 Impact calculation tools, such as github.io, to compute carbon emissions down to the individual GPU.
Estimate Energy Consumption for Every AI Model Training Activity Training a model occurs in phases. With LLMs as an example, the initial pre-training phase is generally the most computationally intensive step and can take months to complete. This can be followed by several “fine-tuning” iterations, where the model is adjusted and validated using additional training data and often human guidance using techniques such as Reinforcement Learning from Human Feedback (RLHF). As noted earlier, two other carbon metrics – inference and overhead – also need to be tracked, by establishing key performance indicators (KPIs) for each training phase metric, as well as inference and operational overhead. The final AI profit/loss analysis should also include projected costs going forward.
Add ML Model Instrumentation and Increase Transparency
For example, AI has been frequently misapplied within the criminal justice system, due to the CSI Effect. Hyperbole both oversells AI capabilities and overstates the dangers it poses. One root cause for this hyperbole is that AI does not intrinsically provide audit trails to explain its decision processes. For this reason, the data used to build AI models needs additional instrumentation to document the influencing factors behind presented results. For true transparency, a model’s methodology should be explainable to employees and customers, and aligned with the company’s core principles. All too often, a company builds its models with the best of intentions, but due to a lack of instrumentation and measurement, don’t detect when the model is being unfair, or even incorrect. They build a model, run some test data through it, and then “fine tune” the model until the test output starts looking correct. But this self-guided testing can’t reliably detect model bias that remains hidden within the pre-training or fine-tuning data.
“Women are more likely than men to occupy a job associated with less status and pay in the data and AI talent pool, usually within analytics, data preparation and exploration, rather than the more prestigious jobs in engineering and machine learning,” stated Erin Young, Judy Wajcman and Leila Sprejer in the Alan Turing Institute’s “Who are the Women? Mapping the Gender Gap in AI” report.
Locally, it’s important to ensure that data and the associated training computation occur in the most energy-efficient cloud configuration available, using infrastructure specifically tuned to the needs of AI training and inferencing workloads when possible. Globally, organizations must consider the environmental advantages of certain sites – which may be geographically distant, and even more expensive in the short term – but have lower energy costs and reduced environmental impacts.
are women in leadership roles, and only 37% of technology startups have women on their boards of directors.
Google
Apple
The carbon footprint from training the model.
The carbon footprint from running inference with the ML model once it has been deployed.
The carbon footprint from operational overhead: all of the supporting computing hardware (including cloud).
An unfortunate truism is that media coverage of AI tends to exaggerate both its benefits and risks, which gives a false impression of how AI impacts sustainability, both for good and bad. AI misconceptions also may lead to misuse and abuse of its capabilities. Not all AI risks are obvious, such as inherent negative biases against certain economic or populations groups that lie hidden within training data.
In ML, the term "ground truth" refers to the accuracy of the training set's classification.
While it’s common to observe model inaccuracies during testing, model inaccuracies arise from unobserved differences in model performance. In ML, the term "ground truth" refers to the accuracy of the training set's classification. In this case, the model is not obviously doing “worse” on one group compared and may be correctly matching the “ground truth” labels in a test dataset, but is still performing unfairly because it replicates or amplifies existing biases or representational gaps in the dataset. Adding instrumentation both helps detect unobserved bias and provides an audit trail that can be published for transparency to employees and customers.
Examine How and Where Data is Stored
11
12
Interactive learning from policy-dependent human feedback, 6 August 2017, Proceedings of the 34th International Conference on Machine Learning - Volume 70
13
We have to ask about emotional AI and crime, Springer Link Open Access, Published: 05 May 2022
Conclusion
Whether AI delivers net total benefits to environmental sustainability depends on whether model training, inferencing, and ongoing operations’ impact on the environment is closely managed. This requires considering energy use in the AI model planning process, tracking the impact of that energy use throughout a model’s lifetime, and instrumenting models to provide auditable results. While it’s true that AI transparency doesn’t directly contribute to sustainability, it can well head off downstream waste due to incorrect AI results. A useful strategy in minimizing environmental impact is exploiting varied geographic hosting locations for bit data and the AI computational resources using that data. With the advent of at least ten international sustainability governance and reporting standards, companies have the legal backing to require sustainability as part of procurement, putting pressure on vendors to disclose AI energy consumption and improve ML tool chain efficiency. Every enterprise exploiting AI must take care to accurately represent both its cost, and its value, avoiding oversell and acknowledging risks.
HOW INFORMA CHAMPIONS SUSTAINABILITY
Faster to Zero: We aim to reduce emissions from our business and compensate for the remaining emissions through high-quality certified offsets by 2025. For example, events generate 50% of our waste, so we’ve partnered with vendors and customers to cut this by half by the end of 2025. We’re aiming for even more sustainability improvements to become a zero waste and net zero carbon business by 2030. Sustainability inside brands: Informa commits to embed sustainability inside 100% of our brands by 2025, with the goal of helping and promoting the achievement of the UN’s Sustainable Development Goals (SDGs). Of the more than 380 events scored in 2023, more than 80% had made a start on embedding sustainability content inside their products. In furtherance of this, Informa conducted 20 workshops and education sessions run to engage colleagues in “sustainability inside”. Multiplying our impact to disconnected people: “Disconnected people” are those who traditionally have struggled to access our knowledge and networks with our products and customer markets. By concentrating on content accessibility to the differently abled, social equity to underserved populations, and understanding cultural differences in, we can design our products to improve accessibility for those with different levels of resource including visas to travel, internet connectivity, and buying power. Environmentally responsible events. Informa’s approach to managing our events sustainably includes two core elements: the Fundamentals, a 12-point checklist of key minimum sustainability expectations that all our events aim to address; and the Accelerator, which helps key events develop more in-depth sustainability programmes. We increased the number of Fundamentals completed by 25% in 2022 to 166 events and work towards top 200 events all being in the Fundamentals. By January 2023, we’ve scored more than 312 events in the Fundamentals system, an increase of 135% from 2021 and significantly exceeding our target, making this one of our most successful engagement programmes across the business in 2022. Colleague equity sharing. At Informa we aim to champion the specialist. We are doing this through our colleague value proposition developed by listening to what colleagues most value about working at Informa. One aspect of collegial involvement is ownership – stock equity in Informa itself – that we foster through our ShareMatch program, one of two current Colleague Equity/Share schemes. By moving faster to net zero, embedding sustainability in our brands, reaching out to disconnected people, environmentally responsible events, and collegial equity sharing, Informa has achieved great strides in sustainability, with rapidly increasing progress to our championship goals.
Pay Attention to Both Sides of the AI Coin
GET IN TOUCH
If you would like to explore our resources and work we do on Sustainability you can visit our website: Alternatively drop us a line if you have any questions or would like to find out more about out sustainability initiatives:
At Informa we’ve made sustainability part of our corporate DNA, with efforts focused on five fronts:
Accelerating movement to net zero carbon and waste emissions.
Embedding sustainability inside each of our brands.
Multiplying our impacts to disconnected people by making our content more discoverable.
Holding environmentally responsible events focused on sustainability.
Bringing our colleagues into our equity sharing programme.
Here are some highlights of Informa's efforts on each of these fronts:
SUSTAINABILITY ACTIVITIES
We’re part of something bigger!
We ask that attendees also recycle their badges post-event, in exit bins provided for this purpose, The badge recycling box weighed in at 30.4 pounds – 13.8 kg at the end of the event.
Our venue, Palmer Event Center, focuses on various sustainability initiatives, including recycling, reusing, and composting 50% of all waste that would be going to landfill. Palmer Events Center is also a carbon neutral venue. Catering vendors at the venue are all local to Austin and leftover food after the event will be donated within the local community.
Applied Intelligence Live! Austin took place on September 21-22 at the Palmer Event Center, a venue and location that aligns with our goals on hosting more sustainable events. The venue is powered by 100% renewable electricity, we removed aisle carpet from our expo to reduce waste, and we were also able to utilize the digital signage to minimize one-off printing, encouraging the exhibitors’ sponsors to think about sustainability in their own procurement decisions.
informa.com/sustainability/
sustainability@informa.com
The AI Summit New York provides a sustainable experience. Efforts include: • Use of renewable electricity • Digital signage to minimize one-off printing • The venue – The Javits Center boasts an impressive nearly 7-acre green roof. This oasis serves as a sanctuary for numerous local and migratory bird species, various bat species, and thousands of insects. The green roof has the capacity to absorb up to seven million gallons of storm-water runoff annually while simultaneously mitigating heat gain within the building. • The Javits Center began harvesting their own honey in 2017, yielding over 250 ounces from beehives situated on the green roof. This honey, known as Jacob's Honey, is thoughtfully distributed in small jars to visitors. How can I contribute to ensuring a sustainable event? • Optimize your business travel by arranging meetings with clients onsite at the event. Keep an eye out for your custom referral codes to facilitate these arrangements. • Bring a reusable water bottle – we have fill stations around the event! • Recycle your badge post-event, so please look out for bins when you are finished with yours!
