Cars

Rush Hour! Photo: Davide Ragusa, 2016-01-16. Davide comments: I took this photo in Borgo Lupo, an abandoned village in Sicily, near Caltagirone (province of Catania). A mystical and empty place, where the only inhabitants are animals and shepherds. Here Sicily expresses its best, with breathtaking surrounding landscapes and smells that smell of the real countryside.

What is this post about? Sheep?

It is about artificial intelligence (AI), and the use of chatbots. A robot is a device that automatically performs complicated, often repetitive tasks. Bot is a shortened form of robot. A chatbot (originally, chatterbot) is a robot that uses and pretends to understands human language. ELIZA was an early chatbot implemented by Joseph Weizenbaum (1923 – 2008) from 1964 to 1967. It did so by passing the Turing test developed by Alan Turing (1912 – 1954) in 1950. This test – originally referred to as the imitation game – means that a human interacting with ELIZA will believe that the robot is another person. It is important to understand that ELIZA and other chatbots do not actually understand English (or any other human language). They store words, then use these and other words to mimic it.

The photo of the sheep was found on Unsplash, a website that allows photos to be freely used, when I was searching for a photo of a traffic jam for the beginning of the post. In much the same way that AI can get things wrong, my use of this photo gets things wrong too. It shows traffic congestion, but with sheep, rather than cars.

Why isn’t the post called Artificial intelligence, and the use of chatbots?

Because, if I gave it that title nobody I know would look at it, let alone read it. Such a title would be offensive to the people I interact with. The people I hang out with are not AI experts.

Why is it called Cars?

An honest answers it is that this weblog’s target readership probably find cars a topic they can relate to. Thus, they are being encouraged to learn something about AI by reading about something they already have a relationship to. Most of my audience readers have driving licenses, and know something about cars. A large proportion of them have been driving/ owning/ servicing/ repairing/ enhancing/ customizing cars for over fifty years. It is a topic they can relate to, unlike, say, the breeding of Labrador dogs.

Do you have something against dogs?

Let me be tactful, just this once, and say I think dogs deserve a companion who is interested in their well being. Many readers of the weblog post have dogs. I live contentedly without them. However, while writing this post, I did find this article about dogs that impressed me.

How did this post begin?

On 2024-01-04, I read an article about Perplexity in Tech Crunch. It is an AI chatbot. I opened a free account, and asked Perplexity some questions. I then tried to find some content that could act as a control to questions answered using perplexity. On 2024-01-13, I read an article in Newsweek, about why Americans can no longer afford cars. I thought it would be interesting to make up questions, based on the answers supplied in Newsweek and then ask Perplexity the same questions. For example, the first question I asked was:

Q. In USA, how much have new and used car prices risen since 2020?

Perplexity provided a long answer, one that answered many different but related questions, rather than just that one. So a new challenge arose about how to present content, so that it made sense. Part of the problem was the attribution of Newsweek content to particular people. I decided to eliminate names and quotation marks. Immediately below is the edited Newsweek answer to that first question.

Since 2020, new car prices have risen by 30 % and used car prices have risen by 38 %.

I was just expecting a simple answer from Perplexity of x% for new, and y% for used vehicles.

Here is more of the Newsweek content, extracted to remove references to sources, human or artificial (Microsoft Copilot).

In 2023—a year during which inflation slowed down to the point that the Federal Reserve decided to stop hiking rates—new car prices rose by 1 percent to an average of $50,364, while used car prices fell by only 2 percent to an average of $31,030.

But as things stand, cars are still really expensive for many Americans. Just 10 percent of new car listings are currently priced below $30,000, Things are not much better in the used car market, where only 28 percent of listings are currently priced below $20,000.

In November 2019, the average transaction price for a new vehicle was $38,500. In November of 2023, that figure jumped to $47,939.

The pandemic’s disruption of manufacturing supply chains, as well as outsized consumer demand in 2021, really put a strain on vehicle inventory and drove prices up significantly.

Dealers practically had customers lined up to buy vehicles that weren’t available. Supply chain disruptions also forced manufacturers to prioritize more profitable, higher-trim vehicles in their lineups, which meant the inventory available to purchase also carried a higher sticker price.

Used car prices will continue to fall modestly in 2024.

The backlog in new and used car demand that grew during the pandemic is slowly moderating, but with over 2 years of restricted new car production, in 2020 through 2022, it will take at least that long for supply to catch up.

Macroeconomic factors like inflation and higher interest rates are also reducing vehicle demand, but not enough to drastically drop car prices in the foreseeable future.

But there are other reasons besides pandemic-related disruptions that there seem to be no more affordable cars in the U.S.—including that automakers are increasingly focusing on the production of expensive SUVs and trucks while dropping smaller, cheaper vehicles that would cost $20,000 or less.

Manufacturers cite disappointing sales results as primary reasons for discontinuing smaller, more affordable vehicles from their lineup.

But car buyers’ preferences have also shifted dramatically to larger trucks and SUVs in the past 10 years or so, and even more towards high-tech and comfort amenities in the form of cameras, sensors, radars and large infotainment screens.

Unfortunately, all these features come at a significantly higher price—even if that’s often higher than the average yearly wage of millions of Americans.

(End of Newsweek content)

Full reply from Perplexity.

According to various sources, new and used car prices in the USA have risen significantly since 2020. The average transaction price for new cars declined 1.4% year-over-year as of October 2023, but they have increased to an average price of $47,936, which is roughly a 60% increase over the last decade1. In July 2020, new-vehicle prices increased $749 (up 2%) from July 2019, while decreasing $473 (down 1.2%) from the previous month2. The average new car price paid in December 2020 and the fourth quarter of last year crossed over $40,000 for the first time, and the average new car price in the US in December 2021 was $47,07734. According to The Zebra, the average new car prices spiked by 17.2% between 2020 and 2021, and the average car payment for new vehicles rose 11.8% to $6445.

Used car prices in the US have also increased significantly since 2020. According to various sources, the prices of used cars have risen due to factors such as the global microchip shortage, supply chain disruptions, and changing consumer preferences

Here are some key points regarding the changes in used car prices:

  • In June 2023, consumer prices for used autos were down 0.5% compared to the previous month and up 5.2% year over year1.
  • The share of used vehicles that sold for less than $20,000 in the first quarter of 2023 was 30.6%, compared to 60.5% just five years ago2.
  • Used car prices have dropped 5.6% compared to a year ago, but they remain higher than at any time before 20203.
  • Used electric vehicle (EV) prices have dropped 26% over the last year3.
  • Car ownership costs, including used car prices, have skyrocketed, with an overall increase of 40% in used vehicle prices between 2020 and 2023.

In summary, used car prices in the US have risen substantially since 2020, and the trend shows no signs of reversing. The factors contributing to the increase include supply chain disruptions, the global microchip shortage, and changing consumer preferences.

According to Statista, the average list price for used vehicles in the United States was around $27,300 in June 2023

Consumer Reports reported that the average price of a used car was around $27,000 in August 20232. Edmunds reported that the average used vehicle transaction price in Q1 2023 dipped 6.4% year-over-year but is still up 44% from 2019, and the share of used vehicles sold for under $20,000 was 30.6% in Q1 20233. The Zebra reported that the average new car prices spiked by 17.2% between 2020 and 2021, and the average car payment for new vehicles rose 11.8% to $6445. Therefore, the average price of a used car in the US in 2023 is around $27,000 to $27,300.

Used car prices in the US are relatively high compared to some other countries. For example, in Germany, the prices of certain car models can be significantly higher than in the US due to factors such as taxes and fees. In the US, the average price of a used car was around $27,000 in 2023. This is in contrast to some other countries where car prices can be influenced by various factors such as taxes, import duties, and government policies. Therefore, while used car prices in the US are high, the comparison to other countries is influenced by a range of economic and regulatory factors.

Chatbots

A human brain has about 86 billion neurons. Chatbots are based on neural networks, imitating the organization of the human brain. An advanced intelligent processing unit (IPU) consists of about 60 billion transistors. A transistor is, at best, an inferior neuron. Since 2017, most modern chatbots are transformer models, a type of machine learning.

If, at this point, I had decided to explain the details of transformer computer architecture, most readers would be putting down their coffee cups and either heading out to their garage workshops, or begging their dogs to leave the comforts of their sofas, and take them for a walk.

While chatbots appear to lie, they are simply responding to input content (often questions). In chatbotspeak, each word a chatbot learns is called a token, and the relationships chatbots construct are called vectors. The tokens and vectors construct a neural network. At first, neural networks use supervised learning to understand relationships between tokens. Here, humans assign labels to the tokens. With unsupervised learning, the neural network itself, without human assistance will assign labels.

Chatbots are designed/ programmed to instill confidence so that users believe they are interacting with a real person. This is the primary goal. Making truthful statements is unimportant, as long as the charade is maintained. A chatbox will do almost anything in order to maintain an illusion of humanness. It will invent information, if that is needed.

Today’s chatbots such as Google’s Bard (now called Gemini – updated 2024-06-15, 22:30), Microsoft’s Copilot, OpenAI’s ChatGPT or the Cohere’s Cohere, use transform technology, first developed in 2017. These are online, generative AI systems that are capable of maintaining a conversation with a user in natural language.

From 1988 to 1991, I taught a college course in AI. Since I had very little faith in machine learning, and chatbots were very primitive, I concentrated on expert systems. To my mind these did the least damage.

Wikipedia tells us: In artificial intelligence, an expert system is a computer system emulating the decision-making ability of a human expert. Expert systems are designed to solve complex problems by reasoning through bodies of knowledge, represented mainly as if–then rules rather than through conventional procedural code. The first expert systems were created in the 1970s and then proliferated in the 1980s. Expert systems were among the first truly successful forms of artificial intelligence (AI) software. An expert system is divided into two subsystems: the inference engine and the knowledge base. The knowledge base represents facts and rules. The inference engine applies the rules to the known facts to deduce new facts. Inference engines can also include explanation and debugging abilities.

If I were wanting to learn about AI today, I would want to start with a fun book. For me, the most enjoyable book on the subject is by Kate Crawford, Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence (2021). Then I would try to read an AI textbook. My first introduction to the topic was: Artificial Intelligence (1983) by Elaine Rich. The most recent edition of that work is a third edition (2009) by Elaine Rich, Kevin Knight and Shivashankar B. Nair. When, about a decade ago, I took an online course in AI with an emphasis on machine learning, the textbook was by Stuart Brand and Peter Norvig, Artificial Intelligence: a Modern Approach. The latest edition is the 4th, from 2020-1. It is much more technical and difficult.

I used Prolog, a computer programming language for expert systems, in my teaching. Initially, I asked my students to map out their family relationships in a knowledge base. Think first of a family with five generations of daughters that would have to be inserted into a knowledgebase: Adriana, Beatrice, Cordelia, Desdemona and Emilia. Then, one would have to make some abstract categories, such as a mother = a female who has a child; a grandmother = a female who has a child who is either the mother of a child or the father of a child. These rules can quickly become very complex. So much of learning Prolog is learning how to create increasingly complex rules.

After students had learned how to systematize family relationships, they tested it, to make sure that the results mirrored reality. A common problem, to begin with, was that grandmothers could only find granddaughters, but not grandsons. Thus, they had to go back and make changes.

Once the family knowedgebase was working, students could go on to work with other problem areas, of their own choosing.

People wanting to learn Prolog as a computing language for an expert system, should probably use William F. Clocksin & Christopher S. Mellish, Programming in Prolog: Using the ISO Standard, 5th edition (2003) as their textbook. This is not as out of date as its publication year would suggest.

Prolog is widely used in research and education. Yet it and other logic programming languages have not had a significant impact on AI. Part of the reason is that most Prolog applications are small and dependent on human experts providing data. Real experts are a scarce resource, and what they know expertly is limited. Thus, few applications exceed 100 000 lines of code.

One Reply to “Cars”

  1. In reality, I typically read ALL your posts … as they are at the least, entertaining, and at the most, informative and thought provoking. So, whether it be cars, sheep in a traffic jam, or AI, I’m pretty much on board!

Leave a Reply

Your email address will not be published. Required fields are marked *