Amazon has joined the swell of concerns regarding the safety of AI, urging thorough testing and robust safeguards before the deployment of more powerful AI models. AI development should not be viewed as a compromise between progress and safety, the company said, suggesting that new systems should be made available to users only after they have been thoroughly tested and deemed safe to use.
The Seattle-based firm, which did not identify itself, said in a statement to Reuters that they “don’t see it as a choice between progress and safety.” “Looking after children begins with robust safety protocols and testing of the models so that they are ready and safe to use.”
Amazon has historically been more muted than some other big tech companies in the public discussion of the potential for long-term dangers from increasingly powerful AI systems, with this announcement being an exception. The firm’s stance coincides with growing worries about safety associated with AI, especially as developers advance the capabilities of models.
The message from Amazon also comes in the wake of some of the top AI companies and executives urging a more balanced approach to the evolution of ever more powerful systems. The calls came on the heels of warnings from researchers and other specialists about the potential severe impacts of advanced AI capabilities if they aren’t properly protected by people or institutions.
Amazon is not advocating for a general slowdown on AI development, but rather highlighting the need to test and ensure safety in addition to advancing technology. It’s important because it’s not just a matter of speeding up AI development anymore in the tech sector. The emphasis is now on how companies can create, and then put into use, increasingly sophisticated systems—and how they can recognize and mitigate potential risks before these systems are released to the public.
“There are risks if that doesn’t happen collectively,” said Amazon, referring to progress in developing AI. We feel, as an industry, in collaboration with the government, we will work through the proper protections,” he said.
Amazon’s strategy puts both technology firms and governments on the hook. The company noted that the industry must work together to develop effective AI safeguards, instead of each laboratory setting its own standards independently. As regulators and policymakers debate the implications of existing regulations on smart artificial intelligence and the need for further safeguards, government intervention may be increasingly relevant.
The challenge is especially complex as AI models can have diverse applications. The same underlying technology can be used for every day consumer services, business applications, software development, and scientific research, and can be used in ways that can lead to misinformation, privacy and cybersecurity risks and other misuses. The models’ capabilities will increase, and testing to identify their traditional technical features might not be sufficient to evaluate whether they are safe to use in the real world.
In the case of companies creating AI algorithms, robust testing may include analyzing the behavior of the models in unusual or intentionally challenging scenarios. Developers can test the model for compliance with safety restrictions, for the generation of misleading information, for its reactions to sensitive requests, for consistency if users try to bypass the safety measures. It is not just a matter of showing that an AI model can work, it is also to discover when and how it may act in an unexpected or harmful manner.
Amazon will have a lot to say since its involvement in AI is not limited to building models. AWS, its cloud computing arm, is hosting infrastructure for business and AI developers. The company also provides customers access to its technology ecosystem, which supports the introduction of AI models from various companies through its Bedrock platform.
This role entails Amazon using parts of the AI industry. It’s also a cloud infrastructure company, an AI technology developer and a platform where companies can get access to models from other companies. The standards are thus important not only for Amazon’s own systems, but for the broader developer and customer community that utilizes Amazon’s infrastructure.
The company builds AI models using AWS, and with a dedicated artificial general intelligence group. This AI work has been applied in its services, such as the ones associated with Alexa and other consumer technologies on Amazon. The company’s investments in AI infrastructure have also positioned it as a key player in the wider growth of generative AI.
The latest statements by Amazon come amid a broader change in the way big tech firms are talking about the responsibility for AI. During the nascent phase of generative AI, the focus of much of the industry discussion was on the rapid development of new capabilities and consumer and business adoption of them. There have been increasing concerns about how systems are tested, their reliability, and transparency and potential misuse, as systems have grown in complexity.
The dispute that is brewing in the industry is partially a speed dispute. Others researchers and technology managers have stated that more careful steps must be taken to build more advanced and robust AI systems, especially if the ramifications of errors are hard to foresee. Others have pointed out that a slowing down of technological advancement could restrict potential applications and give time for competitors to catch up. The situation at Amazon makes the safety testing more important, but doesn’t necessitate a pause in development.
