According to a survey conducted by Gartner among 2,500 executives last spring, nearly 70% of respondents indicated that their organizations are exploring how to integrate generative AI into their operations. A recent Stanford University AI Index Report also highlighted a global increase in adoption rates across various sectors covered in the report.
Technology leader like Microsoft, Meta, Apple, and Salesforce have integrated generative AI into many of their products, offering enterprises the option to create custom versions of large language models (LLMs). Numerous global companies are contemplating the use of generative AI in production, marketing, and beyond.
However, many companies remain hesitant to adopt generative AI applications due to concerns about privacy, security threats, copyright violations, potential bias and discrimination in outputs, and other risks. Some companies have even prohibited their employees from using this technology.
So, what are the risks associated with using this technology, and what are the potential challenges in its adoption and how can they be mitigated?
First: Risks of Generative AI
The risks of generative AI primarily stem from misuse of tools based on this technology and the dissemination of misleading content generated by these tools.
Here are the risks associated with misuse and unchecked dissemination of generative AI:
1. Misuse:
Misuse refers to unethical or illegal exploitation of generative AI capabilities for harmful purposes such as fraud and deceptive campaigns. With advancements in generative AI, fraudsters have used it in various forms of cyberattacks. For instance, the reduced cost of content creation via generative AI has led to an increase in deepfakes, which are used for spreading misinformation, financial fraud, identity theft, and manipulating election results.
2. Propagation of False Information:
A fundamental issue with generative AI is the generation of inaccurate or entirely false outputs, known as 'hallucination'. For example, in June 2023, a radio host in Georgia filed a lawsuit against OpenAI after ChatGPT falsely accused him of fraud and embezzlement from a nonprofit organization.3. Unchecked Content Dissemination:
One of the significant risks associated with generative AI is the use of outputs created by third parties deliberately and disseminating them without verifying their credibility. Recently, numerous deepfake videos and images from unknown sources have been circulated widely, such as a video falsely depicting a Tesla car crash on Reddit in March.
In addition, content unknowingly shared by users without verifying its accuracy can cause significant issues for companies or individuals. For example, users of platform X shared a deepfake image of an explosion at the Pentagon, temporarily affecting stock prices on Wall Street due to mistaken authenticity.
Moreover, the increasing use of deepfakes circumvents security protocols. In 2019, the CEO of a British company fell victim to a fraudulent voice call impersonating the CEO of a major German company, resulting in a €220,000 transfer to the scammer's bank account.
Second: Mitigating Risks of Generative AI
Each of the aforementioned risks presents unique challenges, but there are measures that both public and private sector leaders can take to mitigate these risks.
To mitigate risks associated with misuse of generative AI:
1. Establishing Principles and Guidelines:
Organizations intending to use generative AI should start by establishing clear principles and guidelines to ensure the technology is not used to cause personal or societal harm. These principles should align with the organization’s ethical values.
2. Watermarking AI-generated Content:
When creating content using generative AI, it's crucial to transparently disclose this fact to the public. This ensures transparency and helps in identifying AI-generated content.
To mitigate risks associated with the dissemination of fake content:
1. Awareness in AI Field:
It is essential to establish policies regarding the time allowed for the use of this technology within companies and when it should not be used. Furthermore, training programs should be provided to increase safe and responsible consumption of AI-generated content.
4. Verification of AI Outputs through Specialized Systems:
In addition to watermarking AI-generated content, organizations need to develop advanced systems to verify published content online, reducing the risk of disseminating fake content that may negatively impact a company's reputation.
By implementing these strategies, companies can navigate the complex landscape of generative AI more responsibly, ensuring its use remains beneficial while minimizing potential harms.
