Emerging Directions in AI and Robotics, Explained

UPI Study › Courses › Introduction to Artificial Intelligence › Free lessons › Lesson 9.5 Free with a free account

Emerging Directions in AI and Robotics

Introduction to Artificial Intelligence · Chapter 9: AI Applications, Ethics, and the FutureSection 9.5 · lesson 66 of 6912 min · free with a free account

← AI Across Industries: Manufacturing, Logistics, and BeyondNext lesson: How AI Reshapes Economies and Markets →

Earn 3 college credits for Introduction to Artificial IntelligenceRecommended for credit by NCCRS · Computer Science 373 · self-paced · one payment, lifetime access
Enroll in Introduction to Artificial Intelligence · $250 →Download the syllabus (PDF)

In this lesson

Foundation models, self-driving levels, embodied robots and the AGI debate.

  • Explain what foundation models are and compare prompting, fine-tuning, retrieval and tools for adapting them
  • Classify driving automation using the six SAE levels and explain the long-tail problem
  • Describe multimodal AI, embodied AI, edge computing and the current state of quantum computing for AI
  • Evaluate claims about artificial general intelligence using four critical questions

Free study notes

Get this lesson's notes by email

The key points and 4 practice questions with answers, in your inbox. One email, free.

One email. Unsubscribe in one click.

Transcript

Everything said in the video, section by section (1,582 words). Read along, or search it.

Opening Question

Three Bold Headlines

Sam scrolls past three headlines. Cars now drive themselves. Humanoid robots are coming to every home. Quantum computers will transform AI. Which of these has been shown in a lab, which works reliably in the real world, and which is only a prediction? By the end of this lesson, Sam can sort them, and so can you.

A Useful Habit

Demonstrated, Deployed or Predicted

Earlier lessons covered settled technique: search, learning from data, logic, perception and robot navigation. Here, progress is fast, claims are bold, and many questions are open. So keep one habit. Separate what's been demonstrated in research, what's been deployed reliably, and what's only predicted. Headlines often blur them. And because products come and go, this lesson sticks to ideas that stay useful.

Foundation Models

Trained Once, Reused for Many Tasks

Start with generative AI, which produces new content: text, images, audio, code or video, rather than only labeling or predicting. The big change is foundation models: very large models trained once on broad data, then reused for many different tasks. Earlier systems were built for one job, so each new task meant new data and training from scratch. Now the costly training happens once, and each task needs only a lighter adaptation. Most are pretrained without human labels. A language model learns by predicting the next or missing piece of text. Many image generators learn to turn random noise, step by step, into a picture. And performance often improves fairly regularly as data, computing and model size grow. That observation, called scaling, has steered much of the field.

Adapting a Model

Prompting, Fine-Tuning, Retrieval and Tools

There are four main ways to adapt one. Prompting gives instructions and examples in the input. The model itself is unchanged, so the effort is very low. Fine-tuning continues training on a smaller, specialized data set, changing the model's internal numbers. It suits a style, a field's vocabulary, or a narrow task. Retrieval looks up relevant documents when a question arrives, and hands them to the model with it. Answers can then rest on private or up-to-date documents. And tools let the model call software, run a search or control a machine, often over many steps. Those are the agents that act for a user.

Quick Check

Answering Questions About Weekly Contracts

Quick check. A law firm wants a model to answer questions about its own contracts, and the contracts change every week. Which adaptation fits best? Pause, and decide. Retrieval. When a question arrives, the system looks up the current contracts and hands the relevant ones to the model, so answers rest on this week's documents. Weekly fine-tuning is tempting, but retraining every week is costly, and facts baked into the model's numbers are hard to check or update.

Known Limits

Hallucination, Bias, Cost and Ownership

Foundation models have well-known limits. Hallucination: fluent, confident text that's false, like an invented source. The model generates what's plausible, and doesn't check facts. Bias: it reflects patterns in its training data, including unfair ones. Cost and energy: training and running very large models takes a lot of computing and electricity. Data and ownership: whose work was used, and was privacy respected? And quality is hard to measure: a model can ace a test and still fail where the test didn't look.

Driving Automation

Six SAE Levels From 0 to 5

Now machines that move. For cars, a common scale from an engineering standards body, SAE, has six levels. At Level 0, the human drives, with at most warnings. Level 1 helps with steering or speed. Level 2 handles both, but the human must watch the road at all times. At Level 3, the system drives within certain conditions, but a human must take over when it asks. That's the key line: from here up, the system is doing the driving. Level 4 needs no takeover, but only within its set conditions, like one area. Level 5 drives anywhere a human could.

The Long Tail

Rare Situations Are the Hardest

The hardest part isn't ordinary driving. It's the long tail of rare situations: odd road work, unexpected objects, strange behavior by other drivers, bad weather. A system can handle millions of ordinary kilometers, and still fail on something it has barely seen. Proving it's safe enough, deciding who's responsible when it fails, and writing the rules are open challenges for engineering, law and public trust.

Quick Check

A Driverless Mining Truck's Level

Quick check. A mining truck has no cab and no safety driver. It hauls ore by itself, but only on the mine's own private roads. Which level is it? Pause, and decide. Level four. No human ever has to take over, but it only works within set conditions: the mine's own roads. Five is the trap. No driver doesn't mean anywhere. Level five must handle any road a human could.

Drones and Humanoids

Robots in the Air and in Human Spaces

Drones face the same perception problems in three dimensions, with tight limits on weight, battery and computing. Many navigate with a camera plus inertial sensors, a form of visual SLAM. Safety rules, battery life and flying beyond the operator's sight shape what's practical. Humanoid robots take human form to work in spaces built for people: stairs, doors, tools and shelves. But balance, dexterous hands, limited power and safety near people are all hard. Researchers are trying foundation models for broad robot skills, learning from video and demonstrations. That's still at the research and early-trial stage.

Multimodal and Embodied AI

Many Data Types and Learning Through a Body

Multimodal AI handles several kinds of data in one system. It can take an image and a question and answer in words, describe a video, or turn a spoken request into an action. The world is itself multimodal, and one kind of data can clear up confusion in another. But mistakes can cross between kinds, and quality is hard to measure. Embodied AI learns and acts through a body, like a robot. It runs a loop: perceive, decide, act, and see how the world responds. Real trials are slow, risky and can break equipment, so robots practice in simulation first. The catch is the sim-to-real gap: friction, lighting and countless details differ.

Edge Computing

Running Models on the Device

Edge computing runs AI on the device itself, a phone, a car or a drone, rather than in a remote data center. Three reasons: a car can't wait milliseconds for a network round trip, the connection may be poor or absent, and keeping data on the device protects privacy. The price is limited computing, memory and battery, so models are shrunk. Quantization stores numbers with fewer digits. Pruning removes parts that contribute little. Distillation trains a small model to imitate a large one.

Quantum Computing

Qubits and Unproven Speed-Ups

Quantum computers use qubits, which can hold combinations of zero and one, and can be linked in ways with no everyday counterpart. For a few specific problems, like simulating molecules and materials, that allows algorithms believed to be far faster than any known ordinary method. But today's machines are small and error-prone, and correcting their errors is a major goal. Quantum machine learning is an active research area, with no broad speed-up shown for typical machine learning. So treat claims that quantum computing will soon transform AI as possibilities, not facts.

Open Questions

Reliability, Reasoning and Alignment

Basic questions remain open. Reliability: how can a system say what's true, and admit when it doesn't know? Reasoning: how can long chains of steps avoid piling up errors? Generalization to situations unlike the training data. Learning from a handful of examples, as people do. And whether a system grasps meaning and cause and effect, or only patterns. Interpretability: why did the model answer that? Alignment: does it pursue what its builders and users truly intend? And can capable systems be made far more efficient?

General Intelligence

Three Positions on Reaching AGI

Then there's artificial general intelligence: human-level skill across a wide range of tasks, able to learn new ones. No system is widely accepted as having reached it. One position says scale will be enough: keep growing models, data and computing. The objection: gains may level off, and missing abilities may need new ideas. A second says new ideas are needed, like reliable reasoning and grounding in the physical world. The objection: past claims that something would never be done were often wrong. A third says learning must be combined with structure, knowledge and reasoning. The objection: no one yet knows how to join them in a general way.

Weighing Claims

Four Questions for Any AGI Claim

There's no consensus, partly because the terms are fuzzy. People mean different things, use different tests, and forecast anything from a few years to never. So when you meet an AGI claim, ask four questions. How is the term defined? What evidence is offered? What test would show the claim is wrong? And what does the person making it stand to gain?

Back to Sam

Sorting Sam's Headlines

Back to Sam's headlines. Cars that drive themselves: deployed, but only at some levels and in some conditions. The long tail is still open. Humanoid robots in every home: demonstrated in research and early trials, not deployed. Quantum computers transforming AI: a prediction, with no broad speed-up shown. And for each one, Sam can ask how it's defined, and what evidence is offered.

Key Takeaways

Emerging AI in Summary

To sum up. Foundation models are trained once and adapted by prompting, fine-tuning, retrieval or tools, and they can still hallucinate and carry bias. Driving automation runs from Level 0 to 5, with the big shift between 2 and 3, and the long tail is hardest. Embodied AI practices in simulation, then must close the sim-to-real gap. Edge computing trades power for speed and privacy, and quantum has no proven broad benefit for AI. And the AGI debate has no consensus, so ask how it's defined, what evidence is offered, and what test could prove it wrong.

Think About It

A Robot Claimed to Be Generally Intelligent

Here's something to think about. A company says its new robot is generally intelligent, because it's human-shaped and holds conversations. What would you need to see before believing it? Which of the four questions would you ask first?

About Introduction to Artificial Intelligence at UPI Study

Introduction to Artificial Intelligence (Computer Science 373) is a self-paced online course of 69 video lessons in 9 chapters, from intelligent agents and search to game strategy, neural networks, logic and reasoning, machine learning, language, robotics and AI ethics. It carries a NCCRS credit recommendation for 3 semester credits.

Every lesson transcript is free on pages like this one, and the videos are free with a free UPI Study account. Enrolling adds a practice quiz after each lesson, the graded work and final exam that earn the credit, lifetime access, and an official transcript on completion. UPI Study has helped over 130,000 students with college credits. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.

Enroll now → Course details Download the syllabus All Computer Science courses Every course with PRO →

Compare: AI vs Current Trends in CS and IT

Why take it here

Why students take Introduction to Artificial Intelligence with UPI Study

How machines search, reason and learnIntroduction to Artificial Intelligence covers how AI systems are built: intelligent agents, search and heuristics, game strategy, logic, neural networks, machine learning, language and perception, and the ethics of AI. It is a common upper-level course in computer science degrees.
Credit at a fraction of the priceThe same 3 credits can cost $750 or more at a US college. Here it is one payment, self-paced, with lifetime access.
Finish your degree soonerTransfer credits reduce the courses you still owe, so you can graduate earlier or lighten a heavy semester.
Strengthen an applicationA completed college-level course on an official transcript shows admissions and scholarship committees you can do the work.
Study on your own scheduleNo start dates, no deadlines. Watch the lessons, practice, and take the final when you are ready.
130,000+students helped with college creditsACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.

Where the credit goes

Universities that work with NCCRS

Introduction to Artificial Intelligence carries a NCCRS credit recommendation. These universities work with NCCRS; some are UPI Study's signed partners. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.

Thomas Edison State UniversityThomas Edison State UniversitySigned partnerUniversity of the PeopleUniversity of the PeopleSigned partnerCharter Oak State CollegeCharter Oak State CollegeSigned partnerStrayer UniversityStrayer UniversitySigned partnerColumbia Southern UniversityColumbia Southern UniversitySigned partnerIndiana TechIndiana TechSigned partnerPenn StatePenn StateWorks with NCCRSSouthern New Hampshire UniversitySouthern New Hampshire UniversityWorks with NCCRSSUNYSUNYWorks with NCCRS +many more cooperating universitiesFind my college →

Find my college → Not listed? Ask Athena in the corner of this page, or book a 15-minute call.

FAQ

Questions students ask us

About Emerging Directions in AI and Robotics

What is a foundation model in AI?
A foundation model is a very large model trained once on broad data and then reused for many different tasks. Earlier systems were built for one job, so each new task meant new data and training from scratch. Most foundation models are pretrained without human labels, for example by predicting the next or missing piece of text, and each new task then needs only a lighter adaptation, such as prompting or fine-tuning.
When should you use retrieval instead of fine-tuning a language model?
Retrieval is usually the better fit when answers must rest on private or frequently changing documents. It looks up relevant documents when a question arrives and hands them to the model, so answers reflect the current versions. Fine-tuning continues training on a specialized data set and suits a style, a field's vocabulary or a narrow task, but retraining for every update is costly, and facts stored in the model's numbers are hard to check.
What are the SAE levels of driving automation?
The SAE scale has six levels, numbered 0 to 5. At Level 0 the human drives, Level 1 helps with steering or speed, and Level 2 handles both while the human watches the road at all times. At Level 3 the system drives in certain conditions, but a human must take over when asked. Level 4 needs no takeover within set conditions, such as one area, and Level 5 drives anywhere a human could.
What is edge computing in AI?
Edge computing runs AI on the device itself, such as a phone, car or drone, rather than in a remote data center. It avoids network delays, keeps working with a poor or absent connection, and keeps data on the device for privacy. Because devices have limited computing, memory and battery, models are shrunk with quantization, which stores numbers with fewer digits, pruning, which removes parts that contribute little, and distillation into a smaller model.

About the Introduction to Artificial Intelligence course

How many college credits is Introduction to Artificial Intelligence?
Introduction to Artificial Intelligence (Computer Science 373) carries a NCCRS credit recommendation for 3 semester credits. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.
How much does Introduction to Artificial Intelligence cost and what is included?
$250, paid once, with lifetime access: all 69 video lessons, a practice quiz after each, the graded work and final exam, and an official transcript when you finish. Students who want several courses choose UPI Study PRO instead.
Which universities accept UPI Study credits?
UPI Study has signed partner agreements with universities such as Thomas Edison State University, University of the People and Charter Oak State College, and works with universities such as Penn State, SNHU and SUNY. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities. Use "Find my college" on this page to check yours.
How long does Introduction to Artificial Intelligence take?
It is self-paced with no deadlines. The 69 lessons total about 16 hours of video; most students add reading and practice time and finish in a few weeks to a few months, on their own schedule.
How are the exams done?
Quizzes and assignments are online in the course. The final exam is taken online with proctoring; a person on the UPI Study team checks assignments, usually within two days.
How do my credits reach my college?
When the course shows 100% complete, request the official transcript from your dashboard. It is sent to the college you name, and you can send it to more than one. Check your college first with "Find my college" on this page, or ask Athena.

Ready for the full course? Earn 3 credits in Introduction to Artificial Intelligence.

69 video lessons, a quiz after each one, and the graded work that earns the credit. Recommended for credit by NCCRS.

Enroll in Introduction to Artificial Intelligence →Download the syllabusEvery course with PRO →
Introduction to Artificial Intelligence
3 credits · NCCRSEnroll →