ARTIFICIAL OPERATORS

The potential of artificial intelligence for assisting mill operators is believed to be awesome. By Patrick Whiteway Artificial intelligence (AI) has been compared to a porpoise stranded on a beach — it’s intelligent but it isn’t going anywhere. Many people in the mineral processing industry disagree. About 100 mill designers, operators and managers got together in Ottawa last March to listen to the experts talk about ai and try out some of the latest ai software. Many at the symposium, sponsored by the Canada Centre for Mineral and Energy Technology (canmet), were impressed, if not staggered, by this technology’s immense potential — not only for cutting costs but for making quality improvements, obtaining uniformity of results and making significant gains in milling for relatively little effort. It’s this same optimism and zeal for the new technology that is said to be driving the ai industry worldwide. Optimism is riding high because of the potential markets which pundits see for ai. However, the financial results of some of the computer companies in the industry suggest they are going out of business. In Canada the number of engineers actively working, consulting and selling ai software is so small, compared to other industrialized countries, that some experts argue it isn’t an industry here at all. They say it’s merely a technology being developed elsewhere and that it will take time for it to seep into the Canadian industrial psyche. Growth in demand for ai products, the pundits say, will depend on demand from all industrial sectors rather than just the computer industry. In mining, where flexibility towards change is not institutionalized as it is in the computer industry, demand for new technology has traditionally been slow.

Aids is a growing concern, even in the ai field. But here it stands for Artificial Intelligence Delivery Systems, which are used to accelerate the deployment of ai technology from the university labs, where it is developed, to the industrial sector. As it stands, there are two aids vying for prominence: pc-based systems and more expensive lisp machines. (More on this later.)

In March three major mining companies (Falconbridge Ltd., Inco Ltd. and Noranda Inc.) took the lead in the mining industry by joining 22 other major Canadian companies to form a consortium called Intelligent Systems Inc. Its objective is to pool research funds to finance work on robotics and ai. In this business it pays to be big.

Unisys Inc. is a prime example. This major U.S.-based company emerged last year out of a merger of Sperry, Univac and Burroughs — all sizeable computer companies in their own right. Douglas Broadhurst, president of UNISYS’ Canadian subsidiary, set up shop in Winnipeg, Man., last year. He told delegates at the ai symposium: “We’re not interested in advancing the technology — we’ll leave that up to the universities.” Instead, UNISYS has found a faster way to engineer the use of ai. Its approach is to transfer the knowledge of how ai works to the people who want to use it — people like mill operators. They, like engineers, tend to want to try anything new that comes along in order to improve efficiency. Accountants, on the other hand, want to know if the cost of implementing ai will pay off. So, as is the case with any new technology, managers are left to deal with all the intangibles. They are the ones who will approve mill operators’ requests to attend seminars and courses in ai. Canadian Artificial Intelligence Products Corp. (caip), which is just three years young, is involved with virtually every Canadian university and research organization doing work in ai. It is also dedicated to the practical applications of the technology. This spring caip held one-day seminars in Toronto, Ottawa and Montreal. It has an on-going training program as well.

UNISYS is doing the same sort of thing. Interested companies send representatives to a UNISYS office for one to two weeks of instruction regarding its large-capacity LISP machines; then they return to their own plant for two weeks and return to UNISYS for the final week of training. By then they are well versed in ai technology.

But the representatives have to come prepared. When ai workers (who like to call themselves “knowledge engineers”) get together, the conversation is usually heavy on jargon. Here’s a sampling of some of the issues discussed:

* mum boards (electronic boards that add extra memory space to a computer);

* “getting it up” (i.e. loading the program into a computer’s hard disk);

* heuristics (rules of thumb);

* PROLOG and LISP (two AI programming languages);

* forward and backward chaining (two different reasoning processes);

* inference engines (that part of an ai program that selects the appropriate rules to use and which determines when an acceptable solution has been found);

* the monkey-and-bananas problem (a classic ai problem that has become a kind of industry yardstick);

* and a lot of gibberish about GURU, APES, Insite, Arity, gc and the latest, Goldwork (all ai software packages designed specifically for pcs).

Although ai is being used today, it isn’t being used by many mining companies. So where are all the success stories? Broadhurst says there are 150 examples of ai being used successfully. But the companies which have implemented the technology know they have a strategic advantage over the competition, so they’re not talking about it. Ai should give mining companies (especially base metals producers like Cominco, Falconbridge, Inco and Noranda) an important edge over competitors in Third World countries. (However, Grant Thomas, president of caip Corp, disagrees. He says there are very few ai success stories, at least not in Canada.)

Mill operators will probably be the first in our industry to use the new technology. Those in the know say that, in 5-6 years, ai technology will help operators do their jobs faster, better and with less drudgery. In fact these computer systems will enable mill operators to answer questions accurately — questions which have hitherto been difficult to even ask under real operating conditions. Experts say it will be like taking a lot of the knowledge that is locked up in books and releasing it where it can be used. One computer guru even suggests it will soon be easier to have an intelligent conversation with a microwave oven than with friends. (We’re not sure if this says more about his optimism for ai than about the quality of his friends.) Fact is, efficiency reduces operating costs, which is the main reason ai is being introduced into some milling operations. During the design stage, ai allows companies to examine many more alternatives than is possible under the normal time allotted. As a result, the design will likely be improved.

A computer can help a mill operator anticipate the day-to-day problems of running a mill as well as help him react efficiently to unanticipated problems. The cost of a major spill in a flotation circuit, for example, could be minimized if it were anticipated using ai. Then again, it is impossible to predict the magnitude of a mill’s problems, and some would argue that the cost of applying ai is therefore uneconomical. Others say ai ca n be profitably applied to crushing, grinding, flotation, thickening, filtering and even refining circuits. Such a system could check for inconsistencies in control variables, make sure control variables are still valid under current operating conditions and present a mill operator with a course of action based on a particular set of rules. The operator could then question the line of reasoning used by the computer to arrive at its conclusion before actually exercising control over a circuit. The more complex the milling circuit, the more possible solutions to a problem the computer can analyse in a given period of time — vastly more than the mill operator is capable of analysing in an emergency situation.

But implementing ai poses problems. Although there are close to 130 milling facilities in Canada (41 are gold mills) only 25% were built or expanded s
ince 1980. About 36% of all gold mills were built before 1965. In addition to many of the mills being old, most are very small. Almost half are rated at below 2,000 tons per day. Perhaps the first applications, then, will come in the big uranium, iron ore and base metals mills (see separate story).

Since many mill operators do not have enough faith in electronic sensors to install them mill-wide, implementing ai into mills could be a slow process — one that will take a big commitment and lots of education. One of the advantages of ai technology over conventional computing is the fact that such a system can infer that a particular ph sensor, for example, is not functioning and will not make a decision based on the faulty information coming from that sensor. Examples of such process control systems operating today can be found in petrochemical plants run by Exxon and Shell. “But major oil companies are often very quiet about these things,” says G. M. Swinkels, a Vancouver-based metallurgical consultant who has investigated these systems. He has been involved in an ai project commissioned by a pulp and paper company in British Columbia. That system is being designed to train operators of a multi-million-dollar piece of equipment. The tutor will enable the operator to be exposed to a variety of problems (the equivalent of many years of actual operating experience) before actually operating the machine. Chalk up another one for AI.

Not only will this new way of computing help human operators do a better job; it promises to help them learn jobs much faster. It also holds the potential for giving mine managers the power to become archivists. It promises to enable a company to store on disk the successful reasoning processes of the good mill operator. In fact this is what “knowledge engineers” are trained to do — find out what rules of thumb the good mill operator uses and structure the information-gathering abilities of the computer in such a way that it can emulate this reasoning process. This, of course, suggests ai has long-term potential as an excellent investment.

Computer gurus who have a vested interest in promoting the use of AI say that, by the turn of the century, we’ll have expert management systems, expert financial systems, expert marketing systems and expert services systems running on computers in our mines and mills. For now, the physical mine and mill operations themselves are where the biggest potential for immediate financial benefits lie in our industry. The manager’s office will have to wait. In fact some managers jokingly proclaim it’s okay to automate everything right up to, but not including, the manager’s office. Others say the manager’s office is the logical place to start — to set a leading example for employees to emulate.

Problem-solving with the aid of a computer on the shop floor, though, appears to have the greatest potential for a fast return on the hefty investments required to implement AI technology. Depending on the computer hardware selected, costs can be anywhere from $2,000 to $150,000. Since many mines and mills already own personal computers, some AI buffs argue that the pc version of AI software is the way to go. That way, if a single workstation (out of 100) goes down, only one percent of the company’s computing power goes down. The major computer companies that market $100,000 LISP workstations disagree. Their systems offer the power and flexibility necessary for the complex job at hand, they say. A recent project undertaken by Canada Cement Lafarge in Montreal illustrates their point. Working with UNISYS, Lafarge has developed a prototype expert process control system for a clinker grinding circuit. The company used a software package called KEE, marketed by IntelliCorp (Mountain View, Calif.) and hardware called the ti-Explorer, a computer manufactured by Texas Instruments (Austin, Tex.). The prototype expert system took three months to develop.

The philosophical mine manager might ask: How can we be sure the knowledge in an expert system is correct? The quick answer is that we’re not sure. But the technology is flexible enough that it can be changed, added to and revised. Ai programs are continually changing and being updated by the companies using them. A lot of the time, operators do not have any particular reason for putting a particular bit of information (be it rule or fact) into an expert system. But chances are that the computer will use it, at some point in the future, to make an inference. There is simply no way of knowing that beforehand. So everything goes in, rules that don’t work are trashed and those that do are kept. In this sense, one researcher has suggested that AI be more appropriately called Artificial Ignorance.

The ease with which programs can be changed opens up a big can of worms. What about employee acceptance? What about sabotage? What about corporate security? What about job security? These, and many more questions, are constantly being asked by researchers in government, industry and labor in many countries. SOURCES

The following is a list of suppliers, consultants and universities actively pursuing Artificial Intelligence in Canada: Applied AT Systems Inc. 4015 Carling Ave. Kanata, Ont. K2K 2A3 (613) 592-0084 CAIP Corp. 106 Colonnade Rd., Suite 220 Ottawa, Ont. K2E 72E (514) 464-5339 Evans Research Corp. 1 Eva Rd., Suite 309 Etobicoke, Ont. M9C 3Z5 (416) 621-8814 Heuristics Search Inc. 385 The West Mall, Suite 257 Toronto, Ont. M9C 1E7 (416) 622-8129 International Artificial Intelligence Inc. 5915 Airport Rd., Suite 200 Mississauga, Ont. L4V 1T1 (416) 671-0647 LISP Canada Inc. 290 Boulevard Richelieu Beloeil, Que. J3G 4G5 (613) 727-0082 Logicware Inc. 5915 Airport Rd., Suite 200 Mississauga, Ont. L4V 1T1 (416) 672-0300 Precarn Associates Inc. –* ADDRESS TO COME *– Symbolics Canada Inc. 5915 Airport Rd. Mississauga, Ont. L4V 1T1 (416) 671-0510 Texas Instruments Box 181153 Austin, Tex. 78718 UNISYS Canada 51 Burmac Rd. Winnipeg, Man. R3C 2P7 (204) 786-8486

The following Canadian universities are involved in ai research: McGill University, Dept. of Electrical Engineering, Montreal, Que. Queen’s University, Dept. of Mining Engineering, Kingston, Ont. University of British Columbia, Dept. of Computer Science and Forest Resource Management, Vancouver, B.C. University of Montreal, Dept. of Linguistics and Philology. University of Toronto, Dept. of Computer Science and Medicine. University of Western Ontario, Dept. of Psychology and Computer Science, Waterloo, Ont.