Two of the leading musical manufacturers back then were a California company called Sequential Circuits (who were making the best-selling synthesizer at the time -- early 1980's -- called the Prophet 5) and Roland, a japanese company who had been making popular musical keyboards for awhile, and were at the forefront of making musical interfaces for computers.
Indeed, Roland was one of the first companies to start making musical products that attached to computers. At the time, IBM had just released its first personal computer, the IBM PC, which gave a "serious endorsement" to personal computers. Also, Commodore had released the Commodore 64, one of the first really affordable "home computers", and sales were really starting to take off with personal computers. The Commodore 64 had a built-in analog synthesizer chip made by a musical company called Ensoniq. It wasn't very fancy, but nonetheless, musicians started playing around with it, and quickly began to discover that the programmability of digital computers, combined with a musical instrument, offered them a lot of potential solutions to those problems they were having back then.
Roland saw the potential musical use that computers offered. So Roland began work on a musical interface for the IBM PC. (Unlike the C-64, the IBM PC had no "fancy" built-in sound chip, so there was a void to be filled with a third party product). Roland envisioned a "digital sequencer" to replace the analog sequencers of the time, and it would be built around an IBM PC, which offered a lot of programmability/versatility using tools made by many other companies. Of course, since the PC had no built-in sound chip, and also, Roland wanted this digital sequencer to be able to work with Roland's entire line of new keyboards, Roland decided to build a hardware "musical interface" for the PC. Roland chose to make this an ISA card that plugged into one of the slots of an AT style IBM PC computer, to which an external box was attached that contained further circuitry. It had a lot of intelligent electronics in it, which primarily were designed to add features that turned a computer into a versatile sequencer. For example, it had a built-in metronome and also a tape sync jack so that the computer playback could be synced to a magnetic analog tape recorder such as a reel-to-reel or cassette recorder. It had built-in hardware timers to control the playback, and some filtering options of the digital data flowing between the computer and an external musical instrument. In designing this proprietary PC musical interface, Roland came up with a simple hardware circuit, and a fancy new, "digital language" that they planned to use with all of their upcoming musical keyboards. This digital language would allow the computer and musical instruments to transfer "control data" between them. This new interface would be known as a "Musical Instrument Digital Interface" or MIDI. (Roland loves anacronyms). And the PC card itself would become the MPU-401 ("Musical Processing Unit, model 401" -- gotta love those anacronyms), the first MIDI interface for a computer. (To this day, the MIDI interfaces built into some modern sound cards still offer hardware compatibility to the original MPU-401 since it became such a widely used MIDI interface for a computer).
Roland and Sequential Circuit representatives used to see each other at NAMM (a business trade show for the music industry), and were talking about how customers were wishing that they had some sequencer that worked with the keyboards from all manufacturers. And Roland said, "You know, we're working on this new peripheral for the IBM PC to turn it into a musical sequencer, using a hardware/software protocol in all of our upcoming keyboards. Would you be interested in supporting this? We could both benefit from it since you make such a popular synth, and we'll be making a musical sequencer for these increasingly popular personal computers". The SQ reps said "Sure. We've got some ideas of our own that we're adding to upcoming synths. Maybe we can incorporate these into one standard between us". So Roland sent some design specs to the SQ guys, who made some suggested changes and additions. They both decided to go ahead and adopt this "standard". Then they thought, "Why not see if we can get some of the other leading musical manufacturers to adopt it, as long as it's not really a proprietary standard anymore?". So, they contacted other popular musical manufacturers such as Yamaha and Oberheim, and got them onboard too. MIDI was perhaps the first true effort at joint development among a large number of musical manufacturers.
The first keyboard on the market with a MIDI interface was the Prophet 600 by Sequential Circuits in 1983. Of course, the Roland MPU-401 appeared shortly thereafter with some PC software from Roland that turned the PC into a musical sequencer. Roland made some MPU-401 programming information available to other parties, and soon, there were other PC programs that supported the MPU-401 interface. (Indeed, those other programs proved to be more popular than Roland's initial MESA software. One of those products was called "Cakewalk", made by a small upstart known as "12 Tone Systems", who later changed their name to "Cakewalk" since that product became synonymous with the company). New Roland keyboards also sported a MIDI interface. Yamaha released the DX-7 later that same year, their first keyboard with a MIDI interface. The DX-7 proved to be a huge hit with musicians. They loved the sound of its new, "FM synthesis", bought it in droves, and started to fool around with this new "MIDI interface" thing. Apple computer made a MIDI interface available for its Macintosh computer, and started promoting the computer in the music market during the mid-to-late 1980's since that was a market completely ignored by their biggest competitor IBM, and one that offered them many potential sales, being that musicians loved the flexibility and ease of use of the new, computer-based "digital sequencers". MIDI took off then. By 1985, virtually every new musical keyboard on the market had a MIDI interface.
Roland and SQ thought "Well, we don't want to be responsible for disseminating new additions to MIDI to every music/computer company that supports MIDI. We have too much other work to do. We don't have the time to prepare and mail documents to everyone else". So, all of the musical manufacturers supporting MIDI agreed to start a new organization called the MIDI Manufacturer's Association (ie, MMA). It would be the duty of this new organization to produce/disseminate the paper documents for the MIDI standard, and be the clearing house for new additions/changes to the specification. Members of the MMA would pay some dues to fund the cost of operating this new organization. (Dues were $50 a year, as I recall). And thus, the MMA was born.
It was through the MMA that new additions to the MIDI specification were channeled. For example, Opcode offered the MIDI File Format specification to the MMA, and thus, everyone started creating MIDI sequencer software that could read/write each others' data files. (Prior to that, each sequencer program had its own proprietary file format, and couldn't read each others' data files). Digidesign offered MIDI Time Code (MTC), and thus a standard was created for syncing the playback of various sequencers. (Prior to that, most sequencers also had proprietary sync protocols, for example, a whole mess of incompatible FSK sync interfaces).
In recent years, the MMA hasn't been doing much of anything except for selling photocopies of specifications that were written years ago, and some people have almost forgotten that it even exists. Consequently, there have been very few changes/additions to MIDI for about the past 5 to 10 years. In fact, there have even been some proprietary things creeping in, such as proprietary redefinitions of MIDI status bytes used by multiple bus parallel and serial computer MIDI interfaces. Things often work in cycles though, so as musicians and manufacturers persue proprietary, solo endeavors and ultimately discover the perils of not persuing standards, then things may swing back the other way, and we may eventually see some problems with MIDI solved in a universal way (such as the fact that a single MIDI bus is limited to 16 channels -- too few to accomodate the newer "multi-timbral" synths that weren't around when MIDI was originally conceived, and also the slow-by-today's-standards 32K serial baud rate of MIDI).