• The AI Pulse
  • Posts
  • šŸ§  Chatbots vs. Agents: From Assistance to Autonomy

šŸ§  Chatbots vs. Agents: From Assistance to Autonomy

PLUS: How Appleā€™s Siri-Powered AI Agents Make Your Life Easier

Welcome back AI prodigies!

In todayā€™s Sunday Special:

  • šŸ”Similarities and Differences

  • āš™ļø3 Pre-Action Characteristics of AI Agents

  • šŸ¦¾3 Post-Action Characteristics of AI Agents

  • šŸ”‘Key Takeaway

Read Time: 7 minutes

šŸŽ“Key Terms

  • Nanobody: A small, artificially designed antibody that can fight against various diseases.

  • Large Language Models (LLMs): AI models pre-trained of vast amounts of data to generate human-like text.

  • Hallucinations: When LLMs present false information as fact, often in a confident or matter-of-fact tone.

  • Agentic Workflow: Specialized AI agents working together to autonomously perform complex tasks.

  • Application Programming Interface (API): Allows multiple software applications to share data, features, and functionality back and forth with each other.

šŸ©ŗ PULSE CHECK

Would you let AI agents plan your entire vacation?

Vote Below to View Live Results

Login or Subscribe to participate in polls.

šŸ”SIMILARITIES AND DIFFERENCES

AI agents are the next frontier of AI developments. While AI agents make the headlines every day, most people arenā€™t sure what they actually are. The most common question we see is, ā€œHow is a conversational chatbot different from an AI agent?ā€

Conversational chatbots like OpenAIā€™s ChatGPT provide us with information to help us make decisions by answering open-ended questions with limited context. However, AI agents can operate autonomously across websites, apps, and data types (e.g., text, image, audio, video, and code). Conversational chatbots enhance marginal productivity, but AI agents eliminate swaths of digital work by automating tasks.

Hereā€™re a few examples of Agentic Workflows:

In an Apple Newsroom Press Release, Apple explained how ā€œApple Intelligenceā€ empowers Siri to take hundreds of actions across apps to automate mundane tasks like compiling a grocery list, managing emails, or crafting text replies. For instance, if a friend texts you their new address in Messages, you can say, ā€œHey Siri, add this new address to his contact card.ā€ Or you could ask Siri, ā€œIs my momā€™s flight landing today?ā€ Siri will find the flight details and cross-reference them with real-time flight-tracking information to offer an arrival time.

Researchers at Stanford University (ā€œStanfordā€) developed ā€œThe Virtual Lab,ā€ a platform that enables a team of five specialized AI agents to collaborate on scientific challenges. This platform successfully designed 92 Nanobody candidates to fight the SARS-CoV-2 Virus, a strain of the Coronavirus (COVID-19).

Microsoft introduced ā€œMagentic-One,ā€ a framework that enables a team of four specialized AI agents to collaborate on complex tasks, such as ordering food from a restaurantā€™s website. ā€œMagentic-Oneā€ could be a game changer for data-driven industries. For example, it could help enterprises analyze customer service data to improve customer support processes.

Salesforce CEO Marc Benioff constantly discusses how the ā€œAI agent revolution is realā€ and as exciting as the cloud computing movement and the mobile device wave. When talking about AIā€™s potential, Benioff said, ā€œIā€™ve never been more excited about anything at Salesforce, maybe in my career.ā€ Heā€™s referring to Salesforceā€™s ā€œAgentforce,ā€ which helps companies build AI agents that work together with humans to drive customer success and enhance existing business workflows. ā€œI think weā€™ll have over a billion AI agents running within the next 12 months,ā€ Benioff added.

So, what exactly makes an AI agent an AI agent? How can they help us with our day-to-day lives? Today, weā€™ll bring AI agents to life by diving into one of the most discussed use cases: planning a trip. Weā€™ve all spent hours trying to find the right place to stay at the right price for the best experience. So, how do AI agents help us plan a trip?

According to programmers, developers, and researchers, AI agents share 6 characteristics:

  • 3 Pre-Action Characteristics: Cognitive abilities like perception, socialization, and long-term memory.

  • 3 Post-Action Characteristics: Taking action through autonomy, reactivity, and proactivity.

āš™ļø3 PRE-ACTION CHARACTERISTICS OF AI AGENTS

1. Perception

In cognitive psychology, which examines how people think, learn, and remember, perception occurs when people take in information through their senses. For example, feeling the texture of a fabric or smelling a freshly baked pie. Similarly, AI agents develop a perception of user queries through processing various forms of data across several sources (e.g., text, image, audio, video, and code). For example, AI agents can utilize reviews on TripAdvisor to suggest highly-rated hotels, leverage Google Maps to determine the most efficient transportation methods, and check OpenTable to find the best restaurants.

2. Long-Term Memories

Once AI agents ingest data, they must create, maintain, and update long-term memories about your preferences. For example, building a comprehensive profile of your travel habits or preferences, such as remembering your love for aisle seats or affinity for guided wine tours.

3. Socialization

To take action, AI agents must engage effectively across various platforms, which includes you, other AI agents, apps, and websites. For example, an AI agent could recognize that youā€™re scrolling through Instagram, inferring you have some free time, and send you a push notification asking, ā€œReady to plan your next trip? During the interaction, the AI agent would maintain a natural tone by combining the strengths of specialized chatbots paired with complex databases, such as Amazonā€™s Lex Chatbot, with LLMs that excel in conversational communication, such as OpenAIā€™s ChatGPT.

šŸ¦¾3 POST-ACTION CHARACTERISTICS OF AI AGENTS

Now What?

Once AI agents process your request and obtain the relevant forms of data across several sources, theyā€™ll be ready to take action. The scope of their action depends on their autonomy, and the sophistication of their actions depends on how reactive and proactive theyā€™re programmed to be.

4. Autonomy

The degree of an AI agentā€™s autonomy depends on their boundaries. Basic AI agents operate within strict constraints, making low-impact decisions. For example, a customer support chatbot provides pre-scripted responses to common customer queries. Moderate AI agents still operate within strict constraints but with more flexibility, making moderate-impact decisions with ā€œlocalā€ consequences. For example, scheduling meetings in Google Meet. Advanced AI agents operate across boundaries and control valuable resources, making high-impact decisions with ā€œglobalā€ consequences. For example, a trading framework that allocates billions of dollars in the Global Stock Market, adjusting strategies in real-time based on economic data.

5. Reactivity

Once an AI agent is socialized and educated (i.e., has perceived information to develop long-term memories), it must respond to changes across various platforms. Basic AI agents might update the trip budget on Google Sheets after booking a flight. Advanced AI agents will continuously monitor, analyze, and adjust multiple data streams. If the flightā€™s delayed, itā€™ll immediately rebook the restaurant reservation, relying on numerous APIs to access apps, databases, and websites.

6. Proactivity

Advanced AI agents would anticipate potential issues and offer relevant suggestions without explicit instructions while still deferring the final decision to you. For example, imagine thereā€™s a political protest blocking the bus route. While continuously monitoring social media, the AI agent flags the trending protest on X (i.e., formerly Twitter), identifying its location. The protest also shows up as a red line on the Google Maps bus route to indicate high traffic. So, the AI agent locates the busā€™s refund policy to cancel your ticket. Then, it creates a scheduled Uber from the hotel to the restaurant. Finally, it explains the situation through a notification: ā€œA protest is blocking your route to the restaurant. Iā€™ve canceled the bus ticket free of charge and booked an Uber on a different route. Select ā€˜Yesā€™ to confirm and ā€˜Noā€™ to cancel.ā€

šŸ”‘KEY TAKEAWAY

If an AI agent possesses all 6 characteristics, itā€™s an advanced AI agent that makes high-impact decisions with ā€œglobalā€ consequences.

Advanced AI agents are likely a few years away. Although automated trip planning is compelling and convenient, Reactive and Proactive abilities introduce novel risks. When conversational chatbots experience hallucinations, it doesnā€™t cause harm unless you act on the information. However, when advanced AI agents experience hallucinations, you could lose money if they misinterpret the bus refund policy or privacy if they post your private address on social media instead of a payment portal. Adoption depends on whether the convenience outweighs the slew of risks.

šŸ“’FINAL NOTE

FEEDBACK

How would you rate todayā€™s email?

It helps us improve the content for you!

Login or Subscribe to participate in polls.

ā¤ļøTAIP Review of The Week

ā€œAs DJ Khaled says, ā€˜Another one!ā€™ Iā€™m learning so much about AI because of you.ā€

-Imani (1ļøāƒ£ šŸ‘Nailed it!)
REFER & EARN

šŸŽ‰Your Friends Learn, You Earn!

You currently have 0 referrals, only 1 away from receiving āš™ļøUltimate Prompt Engineering Guide.

Refer 3 friends to learn how to šŸ‘·ā€ā™€ļøBuild Custom Versions of OpenAIā€™s ChatGPT.

Reply

or to participate.