数据显示，约有55％的美国人通过直接购买股票、持有股票共同基金或退休账户投资股票市场。 在收入超过7.5万美元的家庭中这类人群的比例跃升到88％。 谈及投资，很多人把一切都交给别人，比如股票经纪人或财务顾问。 有趣的是，这个托管者的角色正越来越多地由人工智能系统扮演。
已经有大约1360家对冲基金依靠电脑模型来进行股票交易和其他投资。 这些资金代表了1970亿美元的投资者资产，其流向由电脑代码指定。 这些基金中的大多数代表了传统的“量化”（定量）基金，这类基金使用计算机模型预测股价走势并确定交易。
但是，越来越多的对冲基金正完全被人工智能驱动的交易引擎引导。 这些资金是金融市场上的人工智能先锋。 而且，正如被人工智能驱动商业转型的其他市场一样，这些引擎代表着创新的新投资产品，同时也伴随着新的问题。
投资世界在20世纪90年代就开始关注人造智能。 当时的关注重点是人造神经网络（ANN），即将计算机算法建立在为人类大脑提供动力的连接之后。 ANN可以被认为是如今机器学习系统（通过学习大量数据集来进行自我修改的计算机）的前身。 有些人希望神经网络在20世纪90年代改变交易，但革命的时机并不成熟。 然而，他们的遗产——程序化交易——流传了下来。
如今神经网络是程序化交易系统的功能显著增加的体现。 随着神经网络被用于语音助理和自驾车，该技术已经成为主流。 投资基金希望利用这些日益复杂的系统来实现更快、更智能的交易以及更大的利益。
您是否想过为何内幕交易是非法的？ 因为如果某个投资者的信息不能随时提供给其他投资者，那么这个人就拥有了不公平的优势。 这一点十分重要，因为股价在理论上应该反映关于该公司的所有信息。 如果有人知道了其他参与者不知道的信息，则这一优势将为他创造机会，在交易（买入或卖出）股票时即预期到当该信息变得广为人知时可能发生的价格变动。 如果有人事先知道公司的季度销售额会非常强劲，那么在，该人可以以低于信息公开时市价的价格购买该股票。 内幕交易规则等限制投资者在投资活动中以特殊手段获取信息的规则是强化市场参与者之间公平的一种方式。
如果这些独立系统是可行的，那么有理由想象一个由这样的专业系统组成的单一系统。 尽管这一系统远远超出了当今技术的局限，但至少可以看出，一个系统能够利用无数的专家系统，对驱动市场的关键因素进行完美预测。 根据这一逻辑，这一单一的系统在预测任何可交易资产的价格变动方面都将几乎完美。
完美的人工智能超级投资者还尚未被发明。 但这是快速发展的人工智能所需要的一种思考。 因为这样一个完美的系统如果有存在的机会——为了一个个人或团体的利益，损害其他人——我们需要讨论如何应对这一情况。 幸运的是，关于金融奇点的对话已经开始了。
代表第一组是一家人工智能营运基金的创始人Babak Hodjat，他渴望看到由人工智能完成的交易。 他在接受采访时表示：“我们人类总有犯错的时候，对我来说，依靠这些以人为本的直觉和理由远比依靠数据和统计信息可怕得多。
其他人则不认同那种认为一家公司能在没有竞争对手的情况下也能实现这一进步的想法。Ben Carlson认为： “如果有人发现这种伎俩有效，不仅会有其他资金跟进，其他投资者大量投钱进来，真的很难设想一种不仅仅是被套利的情况。”
也有人认为金融奇点是有益的。 按照这个逻辑，一个完全依赖于逻辑的市场可以达到完美的效率，所有的资产都是正确定价的，根本无需人为干预。 耶鲁大学经济学教授Robert J. Schiller表示，计算机将基于包括未来利润、技术进步和人口变化在内的优化预测来设定价格。
Schiller怀疑，金融奇点即将到来。 他表示，金融奇点本应出现在一个只有理性的市场运行的世界。 但人类是不合理的，一个成功的人工智能世界不得不将我们不可预测的本性纳入考虑范围。
目前，交易算法可以相互伪造获得优势，BBC认为这一行为是非法的，但目前还难以给出足够证明。 交易算法也可以预测某些较慢的计划的下一步，然后进行相应的交易。 随着各个公司为了获得更快的交易时间激烈竞争，较慢的计划可能会产生巨大的功能差距。 随着算法变得更加智能化、功能更强大，金融行业将需要更加智慧的保护措施来开发利用风险。
潜在的危险同样存在。 2012年8月，一个基金的交易计划“跑了”，每分钟亏损1000万美元。 人力团队花了近一个小时的时间来确定和解决问题，该公司在此过程中亏损了4.4亿美元。 两年前，一个算法交易造成了“闪电崩溃”，导致美国股市和未来指数在几分钟内下跌了10％。
Data indicates that about 55 percent of Americans are invested in the stock market?through direct ownership of shares, stock mutual fund holdings, or retirement accounts. That figure jumps to 88 percent for households with income greater than $75,000. When it comes to handling their investments, many people turn everything over to someone else, like a stock broker or financial advisor. What’s interesting is how that “someone” increasingly refers to an artificial intelligence system.
Already, some 1,360 hedge funds rely on computer models to trade stocks and other investments. These funds represent $197 billion dollars of investor money being directed by lines of computer code. Most of these funds represent traditional “quant” (quantitative) funds that use computer models to predict share price movements and determine trades.
But an increasing number of hedge funds are entirely directed by AI-powered trading engines. These funds are at the vanguard of AI in financial markets. And, like many markets where?AI is transforming business as usual, these engines represent innovative new investment products while simultaneously raising new questions.
The growth in AI-directed investing could have radical consequences, especially in a scenario where a single investor or investment fund using proprietary AI is able to secure an unfair advantage over other market actors. Call it “stock market singularity.” And the groundwork for such an occurance has already been laid.
Financial neural networks resurrected
The investment world began looking at artificial intelligence in the 1990s. The focus then was on artificial neural networks (ANN), computer algorithms modeled after the connections that power the human brain. ANN can be thought of as a predecessor to today’s machine learning systems, computers that self-modify by learning from massive data sets. Neural networks were expected by some to transform trading in the 1990s, but the revolution never came. However, their legacy — programmatic trading — lives on.
Programmatic trading is computer-controlled investing using algorithms to perform the tasks of traditional investment professionals, like spotting opportunities, managing risk, and making lightning-fast trading decisions. This approach shifts a lot of decision-making onto computers, but the technology by itself hasn’t given one market actor an unfair gain over the others.
Today’s neural networks represent a significant increase in functionality over programmatic trading systems. The technology is already becoming mainstream as neural networks are used in voice-activated assistants and self-driving cars. Investment funds?want to leverage these increasingly sophisticated systems to achieve faster, smarter trades and better yields.
To see why automated AI trading systems might generate a lot of unprecedented challenges, we need to walk through a few ideas.
On the path to an AI super-investor
Ever wonder why insider trading is illegal? It’s illegal because a single person with information not readily available to other investors has an unfair advantage. This is significant because share prices should, in theory, reflect all of the information available about a company. Having knowledge that other participants don’t have creates the opportunity to trade (buy or sell) shares in anticipation of the price change that happens when that information becomes widely known. If someone knows in advance that a company’s quarterly sales will be unusually strong, that person can buy shares at a price lower than they will trade when the information becomes public. Insider trading and other rules that restrict investment activities of investors with special access to information are one way to enforce fairness among market participants.
Getting back to automated AI systems, we next need to get theoretical. It’s possible to imagine an AI system that’s a perfect predictor of a single financial variable like, say, interest rates. Another system might develop infallible inflation predictions. A third might get really good at predicting earnings growth in a particular industry. And so on.
If these individual systems are possible, it’s within reason to imagine a single system made up of several of these specialist systems. And though it’s far outside the limits of today’s technology, it’s at least plausible that a?single system could make use of countless specialist systems that are near-perfect predictors of all of the key factors that move markets. That single system would, by this logic, be virtually perfect at predicting price changes in any — or all — tradeable assets.
Now picture this AI system in the hands of a single investor.
The invention of the perfect AI super-investor is not exactly here yet. But this is the sort of consideration that rapid advances in artificial intelligence require. Because if there’s a chance such a perfect system could come into existence — for the benefit of one individual or group at the detriment of everyone else — it’s worth having a conversation about what to do about it. Luckily, the financial singularity conversation has already started.
Hedging bets on AI
There are, speaking generally, two responses to the financial singularity question. The first is that it’s not actually possible. The second is that it wouldn’t actually be that bad.
Representing the first group is Babak Hodjat, the founder of one AI trading fund, who is eager to see trading in the hands of AI. “It’s well documented we humans make mistakes,” he told Bloomberg. “For me, it’s scarier to be relying on those human-based intuitions and justifications than relying on purely what the data and statistics are telling you.”
Others dismiss the idea that one company will achieve such advances without competitors close on their heels. “If someone finds a trick that works, not only will other funds latch on to it but other investors will pour money into [it]. It’s really hard to envision a situation where it doesn’t just get arbitraged away,” author Ben Carlson told Wired.
There is also the idea that a financial singularity would be beneficial. By this logic, a market that operates purely on logic could reach perfect efficiency, where all assets are priced correctly with no need for human intervention. Computers would set prices based on optimized projections that include future profits, tech advancements, and demographic shifts, according to Robert J. Schiller, a Yale economics professor.
Schiller is skeptical that a financial singularity lies ahead. He argues that it would have to occur in a world where markets run according to rationality alone. But humans are irrational, and a successful AI would have to account for our unpredictable natures.
A future worth pondering
At present, trading algorithms can fake one another out to gain advantages, which the BBC notes is illegal but difficult to prove. They can also predict a slower program’s next moves and then trade accordingly. With firms competing aggressively to get faster trading times, a slower program could create massive functionality gaps. As algorithms become more intelligent and more powerful, the financial industry will require?ever-smarter safeguardsagainst exploitation and risk.
Then there are the potential glitches. In August 2012, a trading program at one fund “ran amok,” creating losses of $10 million a minute. It took nearly an hour for the human team to identify and solve the problem, and the firm lost $440 million in the process. Two years earlier, an algorithmic trade caused a “flash crash,” in which U.S. share and future indices dropped 10 percent within minutes.
Some say those incidents are telling preludes to disaster. A rogue algorithm at one of the country’s major banks, or a cascading failure in which multiple big banks are derailed by faulty programs, could lead to a catastrophic crash.
Whether the financial singularity will happen — and whether its impact would be positive or negative — remains to be seen. But we should all be paying attention because, as we witnessed?in 2008 with the financial crisis, what happens in the market affects us all.