What is wrong with 2016's Artificial Intelligence and why it can't play complex video game like Doom

[mackmack]

As you have said yourself, so far it's a pipe dream. For the near future, at least, the vast majority of AI in use will be specialized.

It appears that your right. But the important thing is that these tech companies are currently developing this universal AI (aka artificial general intelligence aka human level AI). In the next couple of years we will find out wither they succeed or not. If they don't succeed it means we're in the 80's again and there will be another AI winter.

if their AI can play the call of duty (must be able to pass all levels) its a sure bet they have succeeded.

if their AI can play a new call of duty game without prior training then they definitely succeeded -- using prior knowledge to play a brand new video game.

 

[mackmack]

Are there any technology companies that would like to comment about my first post? Any computer magazine editors like wired or computer world? I would greatly appreciate it if the software companies commented because i would like to engage in a meaningful discussion with them, especially Google, IBM, Microsoft, or Stanford university.

 

[Sarkus]

the watson is an expert AI system that can only play jeopardy.

I know English is not your first language but please try not to put "the" in front of proper nouns:
It isn't "the watson", it is just Watson, the name of the AI.
It isn't "the call of duty", it is simply "Call of Duty"

Secondly, Watson can not "only play Jeopardy". Playing Jeopardy was merely a task it was given, with the AI being in how it interpreted the natural language of the questions, in its generation of hypotheses, and its evidence-based learning. It's these 3 things that mean that Watson is now used in medical care to arrive at treatment recommendations following a patient's diagnosis.

As for the video you posted: I see nothing in there apart from someone typing a few things into the Windows Paint program. Which part is meant to the robot?

 

[Sarkus]

Are there any technology companies that would like to comment about my first post? Any computer magazine editors like wired or computer world? I would greatly appreciate it if the software companies commented because i would like to engage in a meaningful discussion with them, especially Google, IBM, Microsoft, or Stanford university.

Why do you think that technology companies would be visiting a site like this? If you want to get them to comment, send them a paper on it, detailing whatever it is you want to show them, including the patents that you have (so that they're not tempted to copy the work) etc. Contact them directly. Don't just hope that someone from a tech company will be on the same forum.

 

[mackmack]

As for the video you posted: I see nothing in there apart from someone typing a few things into the Windows Paint program. Which part is meant to the robot?

you don't see the importance of the video?? you got a program that can think, generate common sense, understand natural language, and do recursive tasks (everything stanford university talked about for the past 20 years). and as far as i'm concerned i can't build this stuff because of the lack of money and resources.

If any technology company builds a future AI to play halo, the thinking process will look exactly like the video. That is why the video is important.

 

[mackmack]

and about Watson, i'm more impressed with googles AI. The IBM Watson is considered weak AI and classified as weak AI. The reason is because ibm spent 5 years building it to play jeopardy. And guess what, they will spend another 10 years building it to recommend treatments for patients. Google is working on one software that can do everything, while Watson is headed in the opposite direction.

Believe it or not Watson can't even do simple math like addition and subtraction. Their software uses a calculator to do 1+1 = 2. The reason humans can not only do simple math, but really complex calculus problems is that they are doing it the long way. the IBM watson gets stuck when the math equation is unsolvable using a calculator.

Most AI researchers today are aiming for Strong AI.

 

[Sarkus]

you don't see the importance of the video?? you got a program that can think, generate common sense, understand natural language, and do recursive tasks (everything stanford university talked about for the past 20 years). and as far as i'm concerned i can't build this stuff because of the lack of money and resources.
If any technology company builds a future AI to play halo, the thinking process will look exactly like the video. That is why the video is important.

I'm sure you think and believe that, but no, you have a video of someone who can think, understand natural language, and do recursive tasks, showing how they would do it and calling it an AI. You might as well just take a video of a book on human cognitive processes and go "ooh, this video is important because it shows how a universal AI will need to operate".

 

[Sarkus]

and about Watson, i'm more impressed with googles AI. The IBM Watson is considered weak AI and classified as weak AI. The reason is because ibm spent 5 years building it to play jeopardy. And guess what, they will spend another 10 years building it to recommend treatments for patients. Google is working on one software that can do everything, while Watson is headed in the opposite direction.

I think you underestimate Watson - it is recommending treatments, and reportedly 90% of those who use it for such subsequently follow its advice.
It's also being used as a virtual teaching assistant - able to answer questions.

Believe it or not Watson can't even do simple math like addition and subtraction. Their software uses a calculator to do 1+1 = 2. The reason humans can not only do simple math, but really complex calculus problems is that they are doing it the long way. the IBM watson gets stuck when the math equation is unsolvable using a calculator.

Most AI researchers today are aiming for Strong AI.

I know - why repeat what has already been done. Watson was built to improve upon natural language processing and other aspects - not to be a general AI. I am merely pointing out that your casual dismissal of Watson is somewhat unwarranted.

 

[mackmack]

Watson is impressive in terms of being an expert system -- having the ability to dig up mountains of data from unstructured data formats. But basic mundane human skills are absent like the concept of gravity and its ability to use math to solve problems.

look at google's AI. Their AI can do sequence data like video games. Watson can only do static data like Q-A and recommendations. Which do you think is a more powerful AI?

 

[mackmack]

I think you underestimate Watson - it is recommending treatments, and reportedly 90% of those who use it for such subsequently follow its advice.
It's also being used as a virtual teaching assistant - able to answer questions.
I know - why repeat what has already been done. Watson was built to improve upon natural language processing and other aspects - not to be a general AI. I am merely pointing out that your casual dismissal of Watson is somewhat unwarranted.

do you know why it's a stupid idea for ibm to pursue weak AI (aka expert systems)? It's because google recognizes that expert systems only assist in human work, but they can't completely replace a human worker (janitors, service workers, factor workers, mail person, truck drivers, etc).

By pursuing Strong AI they can replace the entire work load of a human employee. If watson was built to drive a car, then what happens if the work schedule includes carrying packages to destination locations? Their programmers would have to tweak their machine learning algorithm for each additional human task, which would be impossible to do if they wanted to replace a human janitor. Wouldn't it be a whole lot better just to build one AI software that can do everything.

 

[billvon]

do you know why it's a stupid idea for ibm to pursue weak AI (aka expert systems)? It's because google recognizes that expert systems only assist in human work, but they can't completely replace a human worker (janitors, service workers, factor workers, mail person, truck drivers, etc).

They have already completely replaced human workers in several roles.

By pursuing Strong AI they can replace the entire work load of a human employee. If watson was built to drive a car, then what happens if the work schedule includes carrying packages to destination locations?

Same thing that happens if there is a driver who is so crippled he can't walk - he will require assistance.

 

[mackmack]

They have already completely replaced human workers in several roles.

Same thing that happens if there is a driver who is so crippled he can't walk - he will require assistance.

there you go again. you need to back up your statements with detailed facts. What human workers were replaced by AI for the past 10 years? the last i checked the US is at full employment and common jobs like restaurant workers, teachers, janitors, and even bus drivers are still employed.

I measure how smart AI are by the unemployment rate. high unemployment = impressive AI low unemployment = same old AI methods

 

[billvon]

there you go again. you need to back up your statements with detailed facts. What human workers were replaced by AI for the past 10 years?

Train operators. Phone operators. Telemarketers. Financial analysts. Ticket agents. Bank tellers. Reporters. Pilots. Accountants. Machinists.

the last i checked the US is at full employment and common jobs like restaurant workers, teachers, janitors, and even bus drivers are still employed.

Yep. Indeed, more of those employed are now in service industries, in part because the jobs they once held (accountants, machinists) are now automated. It's a problem.
===========
Service Sector Surges, Manufacturers Learn to Do More With Less

The number of people employed in service industries hit a new high in March, while manufacturing output held steady with fewer workers.

US News and World Report
By Andrew Soergel | Economy Reporter April 28, 2015, at 6:25 p.m.

Anyone looking for a simple picture of how the American economy has changed over the last few decades need only look at the Labor Department's March employment report, in which service sector employment reached an all-time high and the manufacturing sector shed 1,000 positions.

The report shows service sector employment reached a peak last month at 121.6 million workers, nearly 10 times the amount of people employed in manufacturing jobs. Service employment includes a relatively broad swath of the domestic workforce that includes most retail, information technology and health care industries, among others.

Manufacturers have shed more than 7 million positions since the industry's June 1979 peak – when they employed nearly 19 percent of the U.S. workforce. That number now only stands at 8 percent.

“[Technology] has enormously increased the reliance on service jobs,” Jerry Jasinowski, the former president and CEO of the National Association of Manufacturers, says, noting that the industrial sector “is just not going to generate the same number of manufacturing jobs as it has in the past.”. . .

The vast majority of the domestic manufacturing sector was more productive in 2014 than a year earlier, according to a Labor Department report released Tuesday, as automation allows a thinning industrial workforce to maintain relatively consistent output.
=========

I measure how smart AI are by the unemployment rate. high unemployment = impressive AI low unemployment = same old AI methods

Not quite. Impressive automation = higher economic output = more profits = more money to drive the economy = more employment.

I often measure how smart AI's are by what sort of jobs people have. Are they flying jet fighters, doing your taxes and operating trains? AI's not there yet. Are drones replacing pilots? Are automated airport shuttles replacing train drivers? Is TurboTax doing your taxes for you? Are the people thus outsourced flipping burgers instead? Then AI's are here to stay.

 

[mackmack]

Not quite. Impressive automation = higher economic output = more profits = more money to drive the economy = more employment.

I often measure how smart AI's are by what sort of jobs people have. Are they flying jet fighters, doing your taxes and operating trains? AI's not there yet. Are drones replacing pilots? Are automated airport shuttles replacing train drivers? Is TurboTax doing your taxes for you? Are the people thus outsourced flipping burgers instead? Then AI's are here to stay.

your talking about assistant technology that assist people so that companies can be more productive. what i'm talking about are entire jobs being replaced by robots. for example, replacing workers at restaurants would be impressive.

if you look at the problem, its very simple: mass produce meals for people. Current robots can make a sandwich (pre-programmed or somehow taught by demonstration). The problem is when a robot drops the lettuce on the ground, it doesn't understand that the dirty lettuce can cause humans to get sick. So common sense knowledge plays a vital part in preparing and serving a meal. a 6th grade level common sense is needed to work in a restaurant, which current robots don't have.

There are a whole bunch of AI problems that must be solved before any robot can cook meals in a restaurant. Read the current AI journals from MIT and Stanford university on these AI problems. To make the situation even more impossible the robots have to follow strict rules and guidelines from the US government in terms of customer and food safety. If humans have a hard time remembering and obeying these laws, then what makes anyone believe the current AI/robots can follow these complex rules?

do you know of any self-aware robot that can cook in a restaurant (without any human intervention)?

 

[birch]

the army try to recruit people skilled in playing video games, especially gamers who are skilled in the call of duty or halo. And these gamers turn out to be very effective soldiers.

Maybe snipers but soldiers are different. You have to consider the physical, mental, personality etc of the individual if a soldier is a good fit. Someone artistic bent or doesnt take orders well or structure or questions authority etc isnt going to be a good soldier. My son was number either one or two player online call of duty for sometime and very adept at it but he is far from soldier material.

 

[Michael 345]

Are there any technology companies that would like to comment about my first post? Any computer magazine editors like wired or computer world? I would greatly appreciate it if the software companies commented because i would like to engage in a meaningful discussion with them, especially Google, IBM, Microsoft, or Stanford university.

I would like to know why with all the effort put into raising human intelligence, which on face value would have a healthy head start on AI, the IQ in general has not budged.

 

[billvon]

your talking about assistant technology that assist people so that companies can be more productive. what i'm talking about are entire jobs being replaced by robots.

Both are currently occurring.

How many flight engineers are there nowadays? Brakemen? Scribes? Weavers? Those jobs are largely gone, replaced by automation.

How many jobs are there in service industries? Millions, and the sector is growing rapidly, and absorbing the jobs lost to automation. The jobs are generally not as good (a hotel clerk doesn't make as much as an experienced machinist) so there's some loss in income.

if you look at the problem, its very simple: mass produce meals for people. Current robots can make a sandwich (pre-programmed or somehow taught by demonstration). The problem is when a robot drops the lettuce on the ground, it doesn't understand that the dirty lettuce can cause humans to get sick.

And human chefs DO understand that - and still will often pick it up and use it anyway. An automated factory making sandwiches never will, because it will not be set up to do that (and in many cases cannot physically do that anyway.)

There are a whole bunch of AI problems that must be solved before any robot can cook meals in a restaurant. Read the current AI journals from MIT and Stanford university on these AI problems. To make the situation even more impossible the robots have to follow strict rules and guidelines from the US government in terms of customer and food safety. If humans have a hard time remembering and obeying these laws, then what makes anyone believe the current AI/robots can follow these complex rules?

Sure - and a lot more easily than a human can.

The problem with robot chefs is not "remembering the rules." That's trivial. It is manipulating small floppy objects, something that robots are not very good at.

do you know of any self-aware robot that can cook in a restaurant (without any human intervention)?

http://factor-tech.com/robotics/174...als-at-tap-of-a-button-to-go-on-sale-in-2017/

 

[Dinosaur]

From MackMack Post #1

In 2016 the most advanced AI, using deep learning and reinforcement learning, called AlphaGo defeated the best human Go player. If Google's AI can defeat human's at the game of Go, then why can't it play complex video games like call of duty, Halo or Zelda?

I am not familiar with games like Halo or Zelda, but assume that the following remarks are valid.

I assume that computer programs are able to win against humans at Go for the same reason that they were able to beat humans at Chess.

Some chess master (or a team of masters) developed position evaluation algorithms for Chess. I think there were several for use at different stages of the game (not sure of this).

The position evaluation algorithms assign an accurate numeric rating to a given board position.

I assume that there are no known evaluation algorithms for games like Halo or Zelda.​

For chess, the computer initially uses a file of known opening moves. EG: Ruy Lopez or the Spanish opening has been analyzed by master players for some number of moves, after which the player is on his own. (If the opponent deviates from the analyzed moves earlier, the player is on his own earlier).

The computer also has a file of common end games (Example: King & Rook versus King) for use when applicable.

The computer program uses the file of analyzed moves as long as it is applicable, after which it uses the position evaluation algorithms & mini-max logic, described below.

The position evaluation algorithm is used as follows.

If the player has n possible moves, an array with n rows is created, one for each possible move.

Each row contains the a value of the position corresponding to a possible answering move by the computer’s opponent. The computer program assumes that the opponent will choose his best possible move (if not, the computer wins more easily).

The computer program chooses its move by choosing the row for which the best (Id est: the Maximum) opponent reply is the minimum of all the maximums in other rows.​

Hence the term Mini-Max logic.

The position evaluation algorithms used apparently give an accurate rating of any position, and thus allows the use of the computer’s calculating ability to determine the best move at each stage of the game by applying mini-max logic.

I assume there is no method which can assign useful numeric values to a given stage of games like Halo or Zelda.

 

[Dinosaur]

We are along way from computer programs capable of true AI. The chess playing programs might seem like AI, but their programmers have merely discovered methods allowing a computer to win games like chess & Go via brute force computational power.