I am delighted to serve you and honored by your trust
10 Risks of AI and How to Manage Them
AI holds immense value, but achieving the full benefits of AI requires confronting and managing its potential risks. The same advanced systems used in discovering new drugs, screening for diseases, countering climate change, and preserving wildlife and protecting biodiversity can also produce biased algorithms that cause harm and technologies that threaten security, privacy, and even human existence.
Here is a closer look at 10 AI risks and actionable strategies for risk management. Many of the AI risks mentioned here can be mitigated, but AI experts, developers, companies, and governments still need to address them.
Humans are inherently biased, and the AI we develop can reflect these biases. These systems unintentionally learn biases that may exist in training data, appearing in machine learning algorithms and deep learning models that support AI development. These learned biases may persist during AI deployment, leading to biased results.
AI bias can lead to unintended consequences that may be harmful. Examples include applicant tracking systems that discriminate based on gender, health diagnostic systems that provide less accurate results for historically underserved groups, and predictive policing tools that disproportionately target marginalized communities, among others.
Actionable Steps:
Develop an AI governance strategy including frameworks, policies, and processes that guide the responsible development and use of AI technologies.
Establish practices to promote fairness, such as including representative training datasets, forming diverse development teams, integrating fairness metrics, and incorporating human oversight through AI ethics review boards or committees.
Establish processes to reduce bias across the entire AI lifecycle. This includes choosing the right learning model, carefully processing data, and monitoring real-world performance.
Explore AI fairness tools, such as IBM’s open-source AI Fairness 360 toolkit.
Attackers can exploit AI to launch cyberattacks. They manipulate AI tools to clone voices, create fake identities, and craft convincing phishing messages—all with the aim of fraud, hacking, stealing a person's identity, or compromising their privacy and security.
While organizations benefit from technological advances like generative AI, only 24% of generative AI initiatives are secured. This lack of security threatens data exposure and AI model breaches, the global average cost of which reaches USD 4.88 million in 2024.
Actionable Steps:
According to recommendations from the IBM Institute for Business Value (IBM IBV):
Establish a framework for an AI safety and security strategy.
Search for vulnerabilities in AI environments through risk assessment and threat modeling.
Protect AI training data and adopt a security-by-design approach to ensure safe implementation and sustainable development of AI technologies.
Evaluate model vulnerabilities using adversarial testing.
Invest in cyber response training to enhance awareness, preparedness, and security in your organization.
Large Language Models (LLMs) are the foundation AI models for many generative AI applications, such as virtual assistants and chatbots powered by conversational AI. As their name suggests, these language models require a massive volume of training data.
However, data used to train LLMs is often collected via web crawlers that scrape and gather information from websites. This data is often obtained without users' consent and may contain personally identifiable information (PII). Other AI systems providing personalized customer experiences may also collect personal data.
Actionable Steps:
Inform consumers about data collection practices in AI systems: when data is collected, whether it contains specific PII, and how the data is stored and used.
Give users the option to opt out of the data collection process.
Consider using computer-generated synthetic data instead.
AI relies on intensive computational processes that consume significant energy and leave a high carbon footprint. Training algorithms on large datasets and running complex models requires vast amounts of energy, contributing to increased carbon emissions. One study suggests that training a single natural language processing model emits more than 600,000 pounds of carbon dioxide, nearly five times the average emissions of a car over its lifetime.
Water consumption is another concern. Many AI applications run on servers in data centers, which generate significant heat and require large amounts of water for cooling. A study found that training GPT-3 models in Microsoft’s US data centers consumes 5.4 million liters of water, and processing 10 to 50 prompts uses about 500 ml, equivalent to a standard water bottle.
Actionable Steps:
Consider choosing data centers and AI providers that use renewable energy.
Choose energy-efficient AI models or frameworks.
Train models using less data and simplify model architecture.
Reuse existing models and leverage transfer learning, which uses pre-trained models to improve performance on related tasks or datasets.
Consider adopting a serverless architecture and hardware specifically designed for AI workloads.
In March 2023, just 4 months after the launch of ChatGPT by OpenAI, an open letter from tech leaders called for an immediate 6-month pause on training "AI systems more powerful than GPT-4." Two months later, Geoffrey Hinton, known as one of the "Godfathers of AI," warned that the rapid development of AI might soon exceed the level of human intelligence. This was followed by another statement from AI scientists, computer science experts, and prominent figures calling for action to reduce the risk of extinction from AI, considering this risk equivalent to nuclear war and pandemics.
Although these existential risks are considered less immediate compared to other AI-related risks, they remain critically important. Strong AI or Artificial General Intelligence (AGI) refers to a theoretical machine possessing human-like intelligence, while superintelligence refers to a hypothetical advanced AI system that surpasses human intellectual capabilities.
Actionable Steps:
Stay continuously informed on the latest AI research.
Build a strong tech stack while remaining open to experimenting with the latest AI tools.
Upskill AI teams to facilitate the adoption of emerging technologies.
Generative AI has become adept at mimicking creators, producing images reflecting an artist's style, music resembling a singer's voice, or articles and poems similar to a writer's style. However, a major question arises: who owns the copyright for AI-generated content, whether produced entirely by it or with its assistance?
Intellectual property (IP) issues regarding AI-generated works are still evolving, and the ambiguity surrounding ownership rights poses challenges for organizations.
Actionable Steps:
Implement verification mechanisms to ensure compliance with laws regarding licensed works that may be used in training AI models.
Exercise caution when inputting data into algorithms to avoid exposing your company’s IP or others' proprietary information.
Monitor AI model outputs to ensure they are free of content that might reveal your organization’s IP or infringe on others' IP rights.
AI is expected to cause a major transformation in the labor market, raising fears that AI-powered automation will displace workers. According to a World Economic Forum report, nearly half of surveyed organizations expect AI to lead to the creation of new jobs, while nearly a quarter see it potentially causing job losses.
While AI fosters growth in roles like machine learning specialists, robotics engineers, and digital transformation specialists, it simultaneously drives a decline in jobs in other fields. This includes roles such as clerical and secretarial work, data entry, and customer service, to name a few. The best way to mitigate these losses is to adopt a proactive approach that studies how employees can benefit from AI tools to enhance their work, focusing on empowerment rather than replacement.
Actionable Steps:
Reskill and upskill employees to use AI effectively in the short term.
Transform traditional business and operating models, job roles, organizational structures, and other processes to reflect the evolving nature of work.
Create human-machine partnerships that enhance decision-making, problem-solving, and value achievement.
Invest in technology that enables employees to focus on higher-value tasks and drives revenue growth.
One of the more ambiguous and evolving risks in AI is the absence of accountability. Who is responsible when an AI system makes a mistake? Who bears responsibility following harmful decisions resulting from the use of an AI tool?
These questions come to the forefront during fatal incidents and serious collisions involving self-driving cars, or cases of wrongful arrest resulting from facial recognition systems. While these issues are still being worked on by policymakers and regulators, companies can integrate accountability into their AI governance strategy.
Actionable Steps:
Maintain accessible logs and audit trails to facilitate the review of AI system behaviors and decisions.
Keep detailed records of human decisions made during AI design, development, testing, and deployment stages so they can be traced when needed.
Consider using existing frameworks and guidelines that incorporate accountability into AI systems, such as the European Commission's guidelines for trustworthy AI, OECD AI principles, the US National Institute of Standards and Technology (NIST) AI Risk Management Framework, and the US Government Accountability Office AI Accountability Framework.
AI algorithms and models are often viewed as black boxes, where their internal mechanisms and decision-making processes remain obscure, even to AI researchers working closely with the technology. The complexity of AI systems poses challenges when it comes to understanding why they reached a certain conclusion and explaining how they arrived at a specific prediction.
This ambiguity and lack of understanding erode trust and obscure potential AI risks, making it difficult to take proactive measures against them.
Actionable Steps:
Adopt Explainable AI (XAI) techniques. Examples include continuous model evaluation, using Local Interpretable Model-Agnostic Explanations (LIME) to help clarify predictions of classification tools by machine learning algorithms, and Deep Learning Important FeaTures (DeepLIFT) to show traceable links and dependencies between cells in neural networks.
Utilize AI governance through audit and review teams that evaluate the explainability of AI results and define explainability standards.
Explore explainable AI tools, such as IBM’s open-source AI Explainability 360 toolkit.
As with cyberattacks, malicious actors exploit AI technologies to spread disinformation and lies, aiming to influence and manipulate people's decisions and actions. For example, AI-generated robocalls mimicking President Joe Biden's voice were created to persuade a number of US voters not to go to the polls.
In addition to election-related disinformation, AI can produce deepfakes—images or videos modified to fake a person's appearance saying or doing something they never did. These deepfakes can spread via social media, increasing the spread of disinformation, damaging reputations, and exposing victims to harassment or extortion.
AI hallucinations also contribute to the spread of disinformation. These inaccurate but convincing outputs range from simple factual errors to fabricated information that may cause harm.
Actionable Steps:
Educate users and employees on how to recognize disinformation and fakes.
Verify the validity and reliability of information before taking any action based on it.
Use high-quality training data, rigorously test AI models, and continuously evaluate and refine them.
Rely on human oversight to review AI outputs and verify their accuracy.
Keep up with the latest research to detect and combat deepfakes, AI hallucinations, and other forms of disinformation and fakes.
Making AI Governance a Corporate Priority
AI holds great potential, but it also comes with potential risks. Understanding these risks and taking proactive steps to mitigate them can give organizations a competitive advantage.
