Learning-based agents

- [Instructor] Learning based agents do exactly that. They learn from their interactions and improve over time. They use machine learning to adapt and make better decisions in three ways. First, they can use supervised learning where the AI agent is given labeled examples to understand and recognize patterns and make predictions. Second, they can use unsupervised learning before the AI agent explores data and identifies patterns without prior knowledge. And lastly, reinforcement learning uses rewards or penalties so that the AI agent learns to optimize strategies with trial and error. A learning based agent excels at data-driven insights as it can analyze vast amounts of data much faster than a human can. They can also adjust to new situations because they can continuously learn from changing environments. Examples include fraud detection, making personalized recommendations, dynamic pricing, or customer segmentation. These agents also benefit from learning from their mistakes, so look for situations where there are many repeated iterations of similar tasks, where the learning based agent can train itself by observing outcomes over and over again. For example, when making personalized recommendations, does the person choose the options provided to them? But there are also limitations to these agents. Similar to what we've discussed before, data quality matters, but so does data quantity. Biased or insufficient data can result in inaccurate outcomes. And because these AI agents use machine learning and deep neural networks, it's not always transparent how they make decisions. This can be a significant problem for some situations, such as loan approvals, or hiring, where it's essential to detect if there's bias in the decision making. This can be a showstopper for some domains, like financial services or hiring. However, learning-based agents are a good option, if analyzing and acting on vast stories of constantly changing data is at the core of how your organization creates value. If that's the case, look for places where a learning-based agent can do some of the heavy lifting so that you can scale these tasks and grow without adding headcount. Note that training learning-based agents can be intense computationally, while large databases require significant hardware or cloud-based computing resources. We've now reached our last agent type, hierarchical agents. See you in the next video.