Machine Intelligence is technology that enables computers to automatically learn patterns from data and make decisions like a human would - if not better.
Machine intelligence is a set of approaches to automated systems founded on the insight that systems which interact with and explain themselves to humans conversationally, perform as many coordination and data transformation activities as possible 'under the hood,' and are adaptable to a broad set of problem types will realize more value than ad-hoc machine learning analyzes rule-based automated systems.
Let's decompose that dense description. Interact with humans conversationally? Most machine learning techniques require at a minimum the ability to write requests in a programming language to a library of ML algorithms and setting up of a training dataset. An intelligent machine will be able to identify from the sort of question you might ask a colleague what data is required to answer and which of a set
of analytical techniques are necessary to answer it. Just understanding the question and converting it to something machine readable is itself an act of machine learning.
Perform as many tasks 'under the hood' as possible? Every time a human has to take output from one machine learning task and provide it as input to another raises the skill required and complexity of operating a given business process. If the software itself is prepared to identify all steps in a given process and make the handoffs between analytical steps and provide the final output in a user-friendly form, two things happen. First, processes that require some machine learning to be effective become massively scalable, because the data scientist is no longer a bottleneck. Secondly, and relatedly, a wider range of users can interact and benefit from the insights of ML-powered analyzes.
Adaptable to a broad set of problem types? There are many different approaches under the category 'machine learning' ranging from extensions of Bayesian statistics to Deep Learning and Neural Networks. If an integrated system can identify which of these approaches is appropriate to a particular task and provision models trained for it without human selection, than it dramatically increases the set of use-cases for intelligent systems and the scalability of the business processes they support.
To summarize, Machine Intelligence is the transcendence of siloed 'analytical activities' and integration of intelligent, automated techniques into the core of enterprise processes. No longer a technique, something on the way to an arbitrarily scalable colleague.
Often, you will not be able to directly observe 'machine intelligence taking place.' Rather, you will notice improvements in precision, scalability, and adaptability on a given (set of) task(s) that are full stop not possible using traditional data science or expert system approaches.
The more specific, and answerable, form of this question is 'from what are you trying to distinguish Machine Intelligence?' There are two main concrete versions of this: 'How do I know you're not just a team of humans doing the analysis and piping out the results through a software interface, like a twenty-first century Mechanical Turk?' Or 'I accept that this is a fully automated approach, but how do I distinguish an 'intelligent' approach from 'dumb' software?
The answers to each concrete version are actually remarkably similar: Could a single or small team of humans, or a rule-based automated system, come up with results that take into account such a breadth of data, perform its analysis so often/for so many individual cases, or give predictions with such precision? If the answer is no, yet the results exist, then its likely they got there due to a Machine Intelligence.
Of course, for Machine Learning researchers and some elite engineers, they could just ask to see the code. But that's a bit like asking 'how do I know I'm interacting with gravity?' and getting the response 'go do a physics degree and work out the calculations.' For the non-specialist, these questions are more about trusting the recommendations coming out of the system than about understanding the techniques. Thus, the best method is to come up with a list of what allows your organization to trust the recommendations any other human or software approach might make, then when you discover that a broader set of these metrics are more fully satisfied by a new system, that is the best inferential evidence you will have that Machine Intelligence is at work.
MI capabilities fall into three broad categories: make dumb processes interactive and adaptable, replace manual, bespoke processes with ones that scale, and transform directional analyses which miss key mechanisms and opportunities into precise techniques which more comprehensively identify value.
Like the old engineer's cliché quip goes 'that depends.' It depends on the current activities of your team or organization, the data available on which to train models, and the desired state of business processes after the inclusion of Machine Intelligence. With answers to these scoping questions, it will be possible to identify which processes that are currently dumb, rigid, bespoke, and/or notional may become interactive, adaptable, scalable, and/or precise.
In most organizations, there is a tension between high-touch processes which entail a deep understanding of a client's challenges, needs, and options that are very expensive to deploy and traditional automated systems which allow many users to perform some tasks and interactions, but only within the very narrow boundaries of these systems and based on superficial knowledge of the user or client.
Machine Intelligence drastically reduces, and in a few specific cases eliminates, this trade-off. Because the software lives on servers in the cloud, it is always available and can be replicated endlessly and at low cost to provide service to as many users as are logged on to interact with it. Because Machine Intelligent systems can respond to and in natural languages, users can pose questions as they would to a colleague and receive easily understandable answers. Since the machine-learning models powering operations and analyses learn patterns directly from the data, and not from a set of rules, Machine Intelligent processes adapt as conditions change. Drawing on a depth and breadth of data that would be literally impossible for a human analyst with traditional statistical tools to review, Machine Intelligence produces findings that are far more precise and comprehensive than any alternative approach.
If you want your business processes to know more, adapt quickly, scale massively, and interact naturally, your organization should be adopting Machine Intelligence.
First, accept that you have an analytical and/or operational problem. Second, have a method for collecting all irreplaceable internal data relevant to your problem. Lastly, be ready to commit financially and organizationally to transform key functions into data-driven, cybernetic networks.
Integrating Machine Intelligence into your business requires commitment, thoughtfulness, and vision. Even with our standardized products, its not as simple as purchasing a license and pressing play. What will become of the personnel and approaches which previously performed the functions for which Machine Intelligence is being brought in? On what data will the machine-learning models train? How will users be on-boarded and prepared to succeed in the new environment? Answers to all these questions do not need to exist when you begin down this path, but they should have at least been posed.
The most basic problem is the one of data. Machine Intelligence rests on access to relevant data on which to train. Every day, new techniques reduce the number of examples of a given metric necessary to make a defensible prediction and offer more sophisticated ways to deal with missing or messy data. But if you want to take account of the size of the bedrooms in a given apartment or the price of gasoline in a given neighborhood, you will need data on that.
The next question is about processes. Unless you're starting from scratch, whatever operations Machine Intelligence fits into will have points of contact between software and the decision-making. What level of autonomy will you grant the Machine Intelligence? Will you consume the same form of advisory reports just as before but with a different method and source? Or will you redesign the basis for decision-making around the new capabilities? When you're ready to develop answers to process redesign questions, you're ready to incorporate Machine Intelligence.
Lastly, just like in any business, personnel decisions should proceed from an understanding of the roles that need to be filled and the competencies needed to excel in them. Some roles and competencies will change, including being entirely replaced or becoming necessary for the first time, as a result of working with Machine Intelligence. Assessing what will change, how, and what steps your organization should take to successfully make the transition is key to a successful Machine Intelligence roll-out.
We are an enterprise-focused Software as a Service (SaaS) vendor. In addition to the product licenses, we provide configuration and on-boarding support. This can include some data strategy and business process advising, but we are not a consulting or custom development shop.
Machine Colony started with the guiding vision of making the capabilities of Machine Intelligence available to tackle the world's toughest problems without needing to write a line of code, much less hire (or be) a team of fairly elite developers. Obviously, this vision is massively ambitious and will take a much bigger team, and network of collaborators, working for a lot longer to accomplish.
However, we have made significant progress in three key areas. First, is the Spore core library of Machine-Learning architectures and attendant toolset. We have also produced Spire, which is fully-featured software for performing analyses of the direction of real estate markets and the properties within them. Finally, the Spindle data pre-processing software is the most advanced method available for automating away the bane of all data scientists, cleaning and sorting messy data.
So, if you're involved in Real Estate investment or development, we can change your business. If you struggle with data pre-processing challenges regardless of industry, we have something intuitive, scalable and effective for you. For other applications, most centrally to us various activities related to urban development modeling, stay tuned. Great things are coming.
Reach out to us at firstname.lastname@example.org with a short primer (just a few notes is fine) on your challenges and business objectives. Here is an infographic on our conventional engagement cycle. Together, we can discover the additional data, appropriate MI product stack, and process re-engineering necessary to meet your goals.
Seriously, read the engagement cycle doc. And check out our current product offerings. If this looks like something your organization would be interested in, please fill out our inquiry form and we'll get right back to you with some options.