Explore the impacts of recent advancements in artificial intelligence!
Quietly simmering and developing for years, artificial intelligence (AI) entered the global stage with the release of ChatGPT in November 2022. AI permeates many facets of modern society, ranging from production to development and education. While its potential to enhance efficiency and innovation is undeniable, the widespread adoption of AI also raises significant ethical and social concerns.
AI simulates human intelligence by training on massive data sets to develop relationship models. According to Shawn Im, a PhD student working on AI safety with Professor Sharon Li at UW-Madison, AI is meant “to do things humans can’t.”
One of the most prominent benefits of AI lies in its ability to streamline processes and improve efficiency across various sectors. In healthcare, AI-driven diagnostic systems can analyze medical data with speed and accuracy, aiding physicians in disease detection and treatment planning. Similarly, in manufacturing, AI-powered robotics optimize production lines, reducing costs and minimizing errors.
AI also allows the personalization of content through virtual assistants and recommendation systems. Smart home devices, such as Amazon’s Alexa and Google Home, utilize AI algorithms to understand and respond to user commands for tasks like setting reminders or controlling home appliances. Likewise, streaming platforms leverage AI to recommend personalized content based on user preferences.
Recently, machine learning, neural networks, and large-language models (LLMs) have been at the forefront of the AI field. LLMs like OpenAI’s ChatGPT and Google’s Gemini (formerly Bard), while not always entirely accurate, can code, write, and generate human-like responses to questions.
“It’s not something we can fully rely on. As it is now, AI can do more than we thought, which is exciting, but it’s not ready to be used more regularly in the world,” Im remarks.
As such, AI does not currently pose a major threat to jobs. While it may replace humans in several niche cases, it is not reliable enough to be entrusted with a person’s work. Even GPT-4, the most advanced public version of LLMs like ChatGPT, writes buggy code and outputs inaccurate information.
The proliferation of AI raises ethical dilemmas regarding biases in training data. If the model trains on biased data, the model’s output is likely to reflect that bias. The area of work that attempts to mitigate biases in models seeks to align the models with good, human values so it learns not to be harmful.
Another risk AI poses is that of unexpected behavior. In LLMs, unexpected or inaccurate output may not have significant ramifications. Other types of human-interactive models, however, carry more significance in this area.
Autonomous driving, for example, uses AI. Unexpected behavior from a self-driving system in a car could cause a loss of human life. The need for human interaction makes testing for these unexpected or harmful behaviors more difficult because of the risk to the human test subject.
“To understand the level of risk, we need to understand where it is appropriate to use these models.”
Shawn Im
The largest risk AI poses is not posed by AI itself but rather the corporatization of AI. Goals of profit can cause a conflict of interest concerning accuracy, transparency, fairness of information, and harmful biases.
“A safe model does not necessarily mean a beneficial impact. I think it’s necessary that we have ways to enforce standards and regulate how models are developed and used,” Im comments.
The impact of AI on society is multifaceted, encompassing both beneficial advancements and significant risks. While AI has the potential to drive unprecedented progress across various domains, its widespread adoption also raises ethical and social concerns.