Generative AI and Environmental Risks
Introduction
Generative AI models are significant consumers of computational resources and energy required for training and running models. While AI is being hailed as a game-changer, however underneath the shiny exterior, cracks are present which significantly raises concerns for its environmental impact. The development, maintenance, and disposal of AI technology all come with a large carbon footprint. The energy consumption of AI models, particularly large-scale models or image generation systems, these models rely on data centers powered by electricity, often from non-renewable sources, which exacerbates environmental concerns and contributes to substantial carbon emissions.
As AI adoption grows, improving energy efficiency becomes essential. Optimising algorithms, reducing model complexity, and using more efficient hardware can lower the energy footprint of AI systems. Additionally, transitioning to renewable energy sources for data centers can help mitigate their environmental impact. There is a growing need for sustainable AI development, where environmental considerations are integral to model design and deployment.
A breakdown of how generative AI contributes to environmental risks and the pressing need for energy efficiency:
- Gen AI during the training phase has high power consumption, when vast amounts of computational power which is often utilising extensive GPU clusters for weeks or at times even months, consumes a substantial amount of electricity. Post this phase, the inference phase where the deployment of these models takes place for real-time inference, can be energy-extensive especially when we take into account the millions of users of Gen AI.
- The main source of energy used for training and deploying AI models often comes from non-renewable sources which then contribute to the carbon footprint. The data centers where the computations for Gen AI take place are a significant source of carbon emissions if they rely on the use of fossil fuels for their energy needs for the training and deployment of the models. According to a study by MIT, training an AI can produce emissions that are equivalent to around 300 round-trip flights between New York and San Francisco. According to a report by Goldman Sachs, Data Companies will use 8% of US power by 2030, compared to 3% in 2022 as their energy demand grows by 160%.
- The production and disposal of hardware (GPUs, servers) necessary for AI contribute to environmental degradation. Mining for raw materials and disposing of electronic waste (e-waste) are additional environmental concerns. E-waste contains hazardous chemicals, including lead, mercury, and cadmium, that can contaminate soil and water supplies and endanger both human health and the environment.
Efforts by the Industry to reduce the environmental risk posed by Gen AI
There are a few examples of how companies are making efforts to reduce their carbon footprint, reduce energy consumption and overall be more environmentally friendly in the long run. Some of the efforts are as under:
- Google's TPUs in particular the Google Tensor are designed specifically for machine learning tasks and offer a higher performance-per-watt ratio compared to traditional GPUs, leading to more efficient AI computations during the shorter periods requiring peak consumption.
- Researchers at Microsoft, for instance, have developed a so-called “1 bit” architecture that can make LLMs 10 times more energy efficient than the current leading system. This system simplifies the models’ calculations by reducing the values to 0 or 1, slashing power consumption but without sacrificing its performance.
- OpenAI has been working on optimizing the efficiency of its models and exploring ways to reduce the environmental impact of AI and using renewable energy as much as possible including the research into more efficient training methods and model architectures.
Policy Recommendations
We advocate for the sustainable product development process and press the need for Energy Efficiency in AI Models to counter the environmental impact that they have. These improvements would not only be better for the environment but also contribute to the greater and sustainable development of Gen AI. Some suggestions are as follows:
- AI needs to adopt a Climate justice framework which has been informed by a diverse context and perspectives while working in tandem with the UN’s (Sustainable Development Goals) SDGs.
- Working and developing more efficient algorithms that would require less computational power for both training and inference can reduce energy consumption. Designing more energy-efficient hardware, such as specialized AI accelerators and next-generation GPUs, can help mitigate the environmental impact.
- Transitioning to renewable energy sources (solar, wind, hydro) can significantly reduce the carbon footprint associated with AI. The World Economic Forum (WEF) projects that by 2050, the total amount of e-waste generated will have surpassed 120 million metric tonnes.
- Employing techniques like model compression, which reduces the size of AI models without sacrificing performance, can lead to less energy-intensive computations. Optimized models are faster and require less hardware, thus consuming less energy.
- Implementing scattered learning approaches, where models are trained across decentralized devices rather than centralized data centers, can lead to a better distribution of energy load evenly and reduce the overall environmental impact.
- Enhancing the energy efficiency of data centers through better cooling systems, improved energy management practices, and the use of AI for optimizing data center operations can contribute to reduced energy consumption.
Final Words
The UN Sustainable Development Goals (SDGs) are crucial for the AI industry just as other industries as they guide responsible innovation. Aligning AI development with the SDGs will ensure ethical practices, promoting sustainability, equity, and inclusivity. This alignment fosters global trust in AI technologies, encourages investment, and drives solutions to pressing global challenges, such as poverty, education, and climate change, ultimately creating a positive impact on society and the environment. The current state of AI is that it is essentially utilizing enormous power and producing a product not efficiently utilizing the power it gets. AI and its derivatives are stressing the environment in such a manner which if it continues will affect the clean water resources and other non-renewable power generation sources which contributed to the huge carbon footprint of the AI industry as a whole.
References
- https://cio.economictimes.indiatimes.com/news/artificial-intelligence/ais-hunger-for-power-can-be-tamed/111302991
- https://earth.org/the-green-dilemma-can-ai-fulfil-its-potential-without-harming-the-environment/
- https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/
- https://www.scientificamerican.com/article/ais-climate-impact-goes-beyond-its-emissions/
- https://insights.grcglobalgroup.com/the-environmental-impact-of-ai/
Related Blogs
.webp)
Introduction
The spread of misinformation online has become a significant concern, with far-reaching social, political, economic and personal implications. The degree of vulnerability to misinformation differs from person to person, dependent on psychological elements such as personality traits, familial background and digital literacy combined with contextual factors like information source, repetition, emotional content and topic. How to reduce misinformation susceptibility in real-world environments where misinformation is regularly consumed on social media remains an open question. Inoculation theory has been proposed as a way to reduce susceptibility to misinformation by informing people about how they might be misinformed. Psychological inoculation campaigns on social media are effective at improving misinformation resilience at scale.
Prebunking has gained prominence as a means to preemptively build resilience against anticipated exposure to misinformation. This approach, grounded in Inoculation Theory, allows people to analyse and avoid manipulation without prior knowledge of specific misleading content by helping them build generalised resilience. We may draw a parallel here with broad spectrum antibiotics that can be used to fight infections and protect the body against symptoms before one is able to identify the particular pathogen at play.
Inoculation Theory and Prebunking
Inoculation theory is a promising approach to combat misinformation in the digital age. It involves exposing individuals to weakened forms of misinformation before encountering the actual false information. This helps develop resistance and critical thinking skills to identify and counter deceptive content.
Inoculation theory has been established as a robust framework for countering unwanted persuasion and can be applied within the modern context of online misinformation:
- Preemptive Inoculation: Preemptive inoculation entails exposing people to weaker kinds of misinformation before they encounter genuine erroneous information. Individuals can build resistance and critical thinking abilities by being exposed to typical misinformation methods and strategies.
- Technique/logic based Inoculation: Individuals can educate themselves about typical manipulative strategies used in online misinformation, which could be emotionally manipulative language, conspiratorial reasoning, trolling and logical fallacies. Learning to recognise these tactics as indicators of misinformation is an important first step to being able to recognise and reject the same. Through logical reasoning, individuals can recognize such tactics for what they are: attempts to distort the facts or spread misleading information. Individuals who are equipped with the capacity to discern weak arguments and misleading methods may properly evaluate the reliability and validity of information they encounter on the Internet.
- Educational Campaigns: Educational initiatives that increase awareness about misinformation, its consequences, and the tactics used to manipulate information can be useful inoculation tools. These programmes equip individuals with the knowledge and resources they need to distinguish between reputable and fraudulent sources, allowing them to navigate the online information landscape more successfully.
- Interactive Games and Simulations: Online games and simulations, such as ‘Bad News,’ have been created as interactive aids to protect people from misinformation methods. These games immerse users in a virtual world where they may learn about the creation and spread of misinformation, increasing their awareness and critical thinking abilities.
- Joint Efforts: Combining inoculation tactics with other anti-misinformation initiatives, such as accuracy primes, building resilience on social media platforms, and media literacy programmes, can improve the overall efficacy of our attempts to combat misinformation. Expert organisations and people can build a stronger defence against the spread of misleading information by using many actions at the same time.
CyberPeace Policy Recommendations for Tech/Social Media Platforms
Implementation of the Inoculation Theory on social media platforms can be seen as an effective strategy point for building resilience among users and combating misinformation. Tech/social media platforms can develop interactive and engaging content in the form of educational prebunking videos, short animations, infographics, tip sheets, and misinformation simulations. These techniques can be deployed through online games, collaborations with influencers and trusted sources that help design and deploy targeted campaigns whilst also educating netizens about the usefulness of Inoculation Theory so that they can practice critical thinking.
The approach will inspire self-monitoring amongst netizens so that people consume information mindfully. It is a powerful tool in the battle against misinformation because it not only seeks to prevent harm before it occurs, but also actively empowers the target audience. In other words, Inoculation Theory helps build people up, and takes them on a journey of transformation from ‘potential victim’ to ‘warrior’ in the battle against misinformation. Through awareness-building, this approach makes people more aware of their own vulnerabilities and attempts to exploit them so that they can be on the lookout while they read, watch, share and believe the content they receive online.
Widespread adoption of Inoculation Theory may well inspire systemic and technological change that goes beyond individual empowerment: these interventions on social media platforms can be utilized to advance digital tools and algorithms so that such interventions and their impact are amplified. Additionally, social media platforms can explore personalized inoculation strategies, and customized inoculation approaches for different audiences so as to be able to better serve more people. One such elegant solution for social media platforms can be to develop a dedicated prebunking strategy that identifies and targets specific themes and topics that could be potential vectors for misinformation and disinformation. This will come in handy, especially during sensitive and special times such as the ongoing elections where tools and strategies for ‘Election Prebunks’ could be transformational.
Conclusion
Applying Inoculation Theory in the modern context of misinformation can be an effective method of establishing resilience against misinformation, help in developing critical thinking and empower individuals to discern fact from fiction in the digital information landscape. The need of the hour is to prioritize extensive awareness campaigns that encourage critical thinking, educate people about manipulation tactics, and pre-emptively counter false narratives associated with information. Inoculation strategies can help people to build mental amour or mental defenses against malicious content and malintent that they may encounter in the future by learning about it in advance. As they say, forewarned is forearmed.
References
- https://www.science.org/doi/10.1126/sciadv.abo6254
- https://stratcomcoe.org/publications/download/Inoculation-theory-and-Misinformation-FINAL-digital-ISBN-ebbe8.pdf
Introduction
The advent of Electronic Vehicles (EVs) represents a transformative leap towards a more sustainable and environmentally conscious transportation future by nations. However, as these vehicles become increasingly connected and reliant on advanced technological systems, a parallel concern emerges—data privacy. Integrating sophisticated technologies in EVs, such as GPS tracking, biometric authentication, and in-car connectivity, raises substantial questions about the collection, storage, and potential misuse of sensitive personal information. This intersection of automotive innovation and data privacy underscores the need for comprehensive solutions and regulatory frameworks to ensure that the benefits of electric vehicles are realised without compromising the privacy and security of their users.
Electronic vehicles primarily record three types of data;
- Driving behaviour and patterns: The e-vehicle records braking and driving patterns, including acceleration, speed, and swerve. Some vehicles even track air conditioning usage and airbag deployment to determine the point of failure in the event of a crash.
- Location data: The e-vehicles also track GPS systems to gauge the speed and direction of the vehicle.
- EV functions and use of telematic services: Monitoring of EV functions includes battery use management, battery charging history, battery deterioration, electrical system functions and software version information.
Data Privacy requirements of companies
Companies manufacturing e-vehicles are saddled with several data privacy requirements as concerns about consumer safety. Data collected by e-vehicles may be sensitive in nature. Location tracking is a key issue that has garnered attention. The constant recording of a driver's whereabouts can lead to the creation of detailed profiles, raising questions about the potential misuse or unauthorised access to this sensitive information. The risk of surveillance, stalking, or even theft of valuable personal data is a genuine concern for EV owners.
Moreover, integrating smart features, such as voice recognition, biometric authentication, and in-car personal assistants, adds another layer of complexity. These features require the collection and processing of personal data. If not handled securely, they may become vulnerable to hacking or unauthorised access, leading to identity theft or other malicious activities. Additionally, Smart charging systems offer convenience by allowing remote monitoring and control of charging, but they also gather extensive data. The geographical data collected during charging may raise concerns about location privacy.
Striking a delicate balance between leveraging this data for enhancing vehicle performance and user experience while safeguarding the privacy of EV owners is paramount. Transparent privacy policies, secure data storage practices, and stringent encryption protocols are essential components of a comprehensive approach to data protection. If a company is eyeing the international market or utilising cloud-based software with decentralised global data storage, it must also navigate international privacy and data protection laws. A prime example is the General Data Protection Regulation (GDPR), a globally recognised and stringent data protection law applicable to both European-based companies and international entities providing goods, services, or monitoring activities of residents within Europe.
Manufacturers of these vehicles are subjected to compliance with this comprehensive legal framework. Obligations on companies are levied by them being data fiduciaries; dual liability may also emanate since some data fiduciaries may also qualify as data processors. Special care must be taken when data is being transferred to third parties.
Further, compliance with consumer safety laws is also an important consideration. In India, the Consumer Protection Act of 2019 safeguards the rights of consumers, holding manufacturers, sellers, and service providers responsible for any harm resulting from faulty or defective products. This extends the Act's coverage to include manufacturers and sellers of internet and technology-based products. When read with the Digital Personal Data Protection Act of 2023 (DPDP Act), the Consumer Protection Act of 2019 takes on additional significance. The DPDP Act, focusing on the security of an individual's digital personal data, introduces provisions such as mandatory consent, purpose limitation, data minimisation, obligatory security measures by organisations, data localisation, and enforcing accountability and compliance. These provisions apply to information generated by and for consumers, offering a comprehensive framework for protecting digital personal data.
Conclusion
The intersection of e-vehicles and data privacy necessitates a careful and comprehensive approach to ensure the coexistence of automotive innovation and user security. As electric vehicles record intricate data related to driving behaviour, location, and telematic services, companies manufacturing these vehicles must navigate a complex landscape of data privacy requirements. The potential risks associated with location tracking, smart features, and the extensive data collected during charging underscore the importance of transparent privacy policies, secure data storage practices, and stringent encryption protocols. Moreover, as companies expand globally, compliance with international privacy laws like the GDPR becomes imperative. Balancing the enhancement of vehicle performance and user experience with the safeguarding of privacy is paramount. Manufacturers, deemed as data fiduciaries, must exercise diligence, especially when transferring data to third parties. Additionally, adherence to consumer safety laws, such as the Consumer Protection Act of 2019, further emphasises the need for a holistic and vigilant approach to ensure the responsible use of data in the evolving landscape of e-vehicles.
References
- https://digitalcommons.law.scu.edu/cgi/viewcontent.cgi?article=1556&context=chtlj
- https://cyberswitching.com/electric-car-charging-and-data-privacy/#:~:text=Smart%20charging%20systems%20provide%20convenience,in%20safeguarding%20EV%20user%20privacy
Introduction
Assisted Reproductive Technology (“ART”) refers to a diverse set of medical procedures designed to aid individuals or couples in achieving pregnancy when conventional methods are unsuccessful. This umbrella term encompasses various fertility treatments, including in vitro fertilization (IVF), intrauterine insemination (IUI), and gamete and embryo manipulation. ART procedures involve the manipulation of both male and female reproductive components to facilitate conception.
The dynamic landscape of data flows within the healthcare sector, notably in the realm of ART, demands a nuanced understanding of the complex interplay between privacy regulations and medical practices. In this context, the Information Technology (Reasonable Security Practices And Procedures And Sensitive Personal Data Or Information) Rules, 2011, play a pivotal role, designating health information as "sensitive personal data or information" and underscoring the importance of safeguarding individuals' privacy. This sensitivity is particularly pronounced in the ART sector, where an array of personal data, ranging from medical records to genetic information, is collected and processed. The recent Assisted Reproductive Technology (Regulation) Act, 2021, in conjunction with the Digital Personal Data Protection Act, 2023, establishes a framework for the regulation of ART clinics and banks, presenting a layered approach to data protection.
A note on data generated by ART
Data flows in any sector are scarcely uniform and often not easily classified under straight-jacket categories. Consequently, mapping and identifying data and its types become pivotal. It is believed that most data flows in the healthcare sector are highly sensitive and personal in nature, which may severely compromise the privacy and safety of an individual if breached. The Information Technology (Reasonable Security Practices And Procedures And Sensitive Personal Data Or Information) Rules, 2011 (“SPDI Rules”) categorizes any information pertaining to physical, physiological, mental conditions or medical records and history as “sensitive personal data or information”; this definition is broad enough to encompass any data collected by any ART facility or equipment. These include any information collected during the screening of patients, pertaining to ovulation and menstrual cycles, follicle and sperm count, ultrasound results, blood work etc. It also includes pre-implantation genetic testing on embryos to detect any genetic abnormality.
But data flows extend beyond mere medical procedures and technology. Health data also involves any medical procedures undertaken, the amount of medicine and drugs administered during any procedure, its resultant side effects, recovery etc. Any processing of the above-mentioned information, in turn, may generate more personal data points relating to an individual’s political affiliations, race, ethnicity, genetic data such as biometrics and DNA etc.; It is seen that different ethnicities and races react differently to the same/similar medication and have different propensities to genetic diseases. Further, it is to be noted that data is not only collected by professionals but also by intelligent equipment like AI which may be employed by any facility to render their service. Additionally, dissemination of information under exceptional circumstances (e.g. medical emergency) also affects how data may be classified. Considerations are further nuanced when the fundamental right to identity of a child conceived and born via ART may be in conflict with the fundamental right to privacy of a donor to remain anonymous.
Intersection of Privacy laws and ART laws:
In India, ART technology is regulated by the Assisted Reproductive Technology (Regulation) Act, 2021 (“ART Act”). With this, the Union aims to regulate and supervise assisted reproductive technology clinics and ART banks, prevent misuse and ensure safe and ethical practice of assisted reproductive technology services. When read with the Digital Personal Data Protection Act, 2023 (“DPDP Act”) and other ancillary guidelines, the two legislations provide some framework regulations for the digital privacy of health-based apps.
The ART Act establishes a National Assisted Reproductive Technology and Surrogacy Registry (“National Registry”) which acts as a central database for all clinics and banks and their nature of services. The Act also establishes a National Assisted Reproductive Technology and Surrogacy Board (“National Board”) under the Surrogacy Act to monitor the implementation of the act and advise the central government on policy matters. It also supervises the functioning of the National Registry, liaises with State Boards and curates a code of conduct for professionals working in ART clinics and banks. Under the DPDP Act, these bodies (i.e. National Board, State Board, ART clinics and banks) are most likely classified as data fiduciaries (primarily clinics and banks), data processors (these may include National Board and State boards) or an amalgamation of both (these include any appropriate authority established under the ART Act for investigation of complaints, suspend or cancellation of registration of clinics etc.) depending on the nature of work undertaken by them. If so classified, then the duties and liabilities of data fiduciaries and processors would necessarily apply to these bodies. As a result, all bodies would necessarily have to adopt Privacy Enhancing Technologies (PETs) and other organizational measures to ensure compliance with privacy laws in place. This may be considered one of the most critical considerations of any ART facility since any data collected by them would be sensitive personal data pertaining to health, regulated by the Information Technology (Reasonable Security Practices And Procedures And Sensitive Personal Data Or Information) Rules, 2011 (“SPDI Rules 2011”). These rules provide for how sensitive personal data or information are to be collected, handled and processed by anyone.
The ART Act independently also provides for the duties of ART clinics and banks in the country. ART clinics and banks are required to inform the commissioning couple/woman of all procedures undertaken and all costs, risks, advantages, and side effects of their selected procedure. It mandatorily ensures that all information collected by such clinics and banks to not informed to anyone except the database established by the National Registry or in cases of medical emergency or on order of court. Data collected by clinics and banks (these include details on donor oocytes, sperm or embryos used or unused) are required to be detailed and must be submitted to the National Registry online. ART banks are also required to collect personal information of donors including name, Aadhar number, address and any other details. By mandating online submission, the ART Act is harmonized with the DPDP Act, which regulates all digital personal data and emphasises free, informed consent.
Conclusion
With the increase in active opt-ins for ART, data privacy becomes a vital consideration for all healthcare facilities and professionals. Safeguard measures are not only required on a corporate level but also on a governmental level. It is to be noted that in the 262 Session of the Rajya Sabha, the Ministry of Electronics and Information Technology reported 165 data breach incidents involving citizen data from January 2018 to October 2023 from the Central Identities Data Repository despite publicly denying. This discovery puts into question the safety and integrity of data that may be submitted to the National Registry database, especially given the type of data (both personal and sensitive information) it aims to collate. At present the ART Act is well supported by the DPDP Act. However, further judicial and legislative deliberations are required to effectively regulate and balance the interests of all stakeholders.
References
- The Information Technology (Reasonable Security Practices And Procedures And Sensitive Personal Data Or Information) Rules, 2011
- Caring for Intimate Data in Fertility Technologies https://dl.acm.org/doi/pdf/10.1145/3411764.3445132
- Digital Personal Data Protection Act, 2023
- https://www.wolterskluwer.com/en/expert-insights/pharmacogenomics-and-race-can-heritage-affect-drug-disposition
